WO2022141538A1

WO2022141538A1 - Trajectory prediction method and apparatus

Info

Publication number: WO2022141538A1
Application number: PCT/CN2020/142439
Authority: WO
Inventors: 凌立; 周伟
Original assignee: 华为技术有限公司
Priority date: 2020-12-31
Filing date: 2020-12-31
Publication date: 2022-07-07
Also published as: CN112839855B; CN112839855A

Abstract

A trajectory prediction method and apparatus. The method comprises: obtaining a first position of a target vehicle at a first moment, the first position corresponding to a first lane set comprising at least one first lane; and then determining a target predicted trajectory, the confidence level of the target predicted trajectory being greater than or equal to a first threshold, the target predicted trajectory being at least one of at least one candidate predicted trajectory, any candidate predicted trajectory comprising a first lane and a second lane that are connected, and a second lane set corresponding to a predicted second position of a target lane at a second moment. By means of the solution, a set of lanes can be determined according to different positions of the target vehicle at different moments, so as to determine at least one candidate predicted trajectory. The present solution can achieve a higher accuracy than the existing methods of only determining a predicted trajectory according to a current position of a vehicle. Then, at least one candidate predicted trajectory having a confidence level greater than the first threshold is used as the target predicted trajectory, such that trajectory prediction accuracy is further improved.

Description

A kind of trajectory prediction method and device

technical field

The present application relates to the field of intelligent driving, in particular to a trajectory prediction method and device.

Background technique

With the development of trajectory prediction technology, the role of trajectory prediction in the field of intelligent driving has become increasingly prominent. Trajectory prediction refers to the prediction of other vehicles in the future in a short period of time after determining the positions of other vehicles around the vehicle through sensors or on-board cameras. The possible driving trajectory, and then planning and controlling its own driving behavior.

At present, when the trajectory prediction of the target vehicle is performed, the heading angle of the target vehicle is determined according to the current position of the target vehicle. According to the current position of the target vehicle and the preset query radius, a lane set including at least one lane is searched in the map, and for each lane in the found lane set, the vertical distance between the current position of the target vehicle and the lane is determined , a candidate lane set including at least one candidate lane is obtained after removing the lanes whose vertical distance is greater than the preset threshold in the lane set. For each candidate lane, determine the projection position of the current position of the target lane on the candidate lane, and determine the angle that the target vehicle needs to turn if the target vehicle travels to the candidate lane according to the projection position and the median line of the candidate lane, and then determine the heading angle. The angle difference between the angle difference and the included angle, according to the order of the angle difference corresponding to the candidate lanes from small to large, select the first N candidate lanes to determine the predicted trajectory, each predicted trajectory contains a candidate lane, where N is greater than or equal to 1. positive integer.

It can be seen from the above introduction that at present, when performing lane prediction, the target vehicle's driving trajectory is only predicted based on the current position of the target vehicle, the vehicle heading angle and other information. However, when the target vehicle is in a complex intersection scene, such as an overpass intersection, it may lead to Inaccurate lanes after projection and wrong predicted trajectories; and when the driver has irregular driving behavior, only predicting the trajectory based on the current position of the vehicle and the vehicle heading angle information may also lead to unreasonable predicted trajectories. The phenomenon. Therefore, the current trajectory prediction methods are not accurate enough.

SUMMARY OF THE INVENTION

The present application provides a trajectory prediction method and device to improve the accuracy of trajectory prediction.

In a first aspect, an embodiment of the present application provides a trajectory prediction method. The method includes:

obtaining a first position of the target vehicle at a first moment, the first position corresponding to a first lane set including at least one first lane; determining a target predicted trajectory, the confidence of the target predicted trajectory is greater than or equal to a first threshold, The target predicted trajectory is at least one of at least one candidate predicted trajectory, and any candidate predicted trajectory in the at least one candidate predicted trajectory includes the first lane in the first lane set and the first lane in the second lane set. Two lanes, the first lane and the second lane are connected in the candidate predicted trajectory; wherein, the second lane set corresponds to a second position, and the second position is the target vehicle at the second moment. A predicted position, the second time instant is temporally behind the first time instant.

In the above method, the first position of the target vehicle corresponds to the first lane set, the second position of the target vehicle corresponds to the second lane set, and the determined at least one candidate predicted trajectory includes the first lane and the second lane in the first lane set. The second lane in the lane set, through this method, it can be determined according to the first lane set corresponding to the first position of the target vehicle at the first moment and the predicted second lane set corresponding to the second position of the target vehicle at the second moment At least one candidate prediction trajectory, to improve the accuracy of the determined candidate prediction trajectory. The target predicted trajectory is at least one candidate predicted trajectory selected from at least one candidate predicted trajectory with a confidence greater than the first threshold. The trajectory prediction method provided in this application uses the confidence to measure the accuracy of the candidate predicted trajectory. At least one candidate predicted trajectory with a threshold is used as the target predicted trajectory, which further improves the accuracy of the predicted target predicted trajectory, and can provide a more credible target predicted trajectory for the downstream modules of trajectory prediction in intelligent driving, such as the planning control module, thereby improving the Intelligent driving performance.

In a possible design, the first lane set is determined according to the first position and first lane information; and/or the second lane set is determined according to the second position and second lane information The first lane information includes the position information of all or part of the lanes in the area to which the first position belongs, and the second lane information includes the position information of all or part of the lanes in the area to which the second position belongs. , the first lane information and the second lane information are the same or different.

Through this design, the first lane set is determined based on the first position of the target vehicle, the second lane set is determined based on the second position of the target vehicle, and the lane sets corresponding to the positions of the target vehicle at different times can be determined.

In a possible design, the first lane information and/or the second lane information is pre-configured or acquired through vehicle communication or through sensors configured in the vehicle. For example, lane information can be pre-configured in the vehicle, or the vehicle can acquire lane information through V2X communication with roadside units or monitoring centers, or detect and perceive lane information in real time through sensors such as on-board cameras and lidars.

Through this design, lane information can be pre-configured in the vehicle, acquired by the vehicle through vehicle communication, or detected in real time, providing multiple ways to acquire lane information.

In a possible design, the method further includes: acquiring the first lane information according to a map of the area to which the first position of the target belongs, and acquiring the second lane information according to a map of the area to which the second position belongs; Taking at least one lane within a preset query radius from the first position as the first set of lanes, and/or taking at least one lane within a preset query radius from the second position as the first lane set Two-lane set.

Through this design, the first lane information can be obtained through the map of the area to which the first position belongs, and the second lane information can be obtained through the map of the area to which the second position belongs. After acquiring the first lane information and the second lane information, the first lane set and/or the second lane set may be determined according to the preset query radius, so as to determine the lane sets corresponding to the positions of the target vehicle at different times.

In a possible design, before the target predicted trajectory is determined, the at least one candidate predicted trajectory is determined according to the first lane set and the second lane set.

Through this design, at least one candidate predicted trajectory is determined according to the first set of lanes and the second set of lanes, and the obtained candidate predicted trajectory includes the lanes that the target vehicle may pass through at the first position and the lanes that the target vehicle may pass through at the second position. lanes, so that the determined candidate predicted trajectories are more accurate.

In a possible design, the determining the at least one candidate predicted trajectory according to the first lane set and the second lane set includes: when the target second lane in the second lane set is the same as the second lane set When the target first lane in the first lane set has connectivity, a target candidate predicted trajectory including the target second lane and the target first lane is determined; wherein the target candidate predicted trajectory is included in the at least one target candidate predicted trajectory. candidate prediction trajectory.

Through this design, the candidate predicted trajectory can be determined according to the first lane and the second lane that has connectivity with the first lane, ensuring that the first lane and the second lane in the obtained candidate predicted trajectory are connected, and the determined candidate predicted trajectory It is the actual trajectory that the vehicle may travel.

In a possible design, the confidence of the target predicted trajectory is obtained according to the first similarity, the second similarity and the connectivity information, where the first similarity is used to indicate the history of the target vehicle The similarity between the trajectory and the first lane in the target predicted trajectory, and the second similarity is used to indicate the difference between the predicted driving trajectory from the first position to the second position and the second lane. The connectivity information is used to indicate the connectivity between the first lane and the second lane.

Through this design, the first similarity between the historical trajectory of the target vehicle and the first lane in the target predicted trajectory is determined, and the second similarity between the predicted driving trajectory and the second lane in the target predicted trajectory is determined. According to the first similarity, The second similarity and the connectivity information between the first lane and the second lane are used to determine the confidence of the target predicted trajectory. The obtained confidence of the target predicted trajectory combines the historical trajectory of the target vehicle and the predicted driving trajectory in two stages. trajectories, and the connectivity information between lanes is introduced, so that the determined confidence level can better reflect the accuracy of the target predicted trajectory.

In a possible design, the confidence of the target predicted trajectory is determined according to the following manner: according to the first similarity, the first weight corresponding to the first similarity, the second similarity, the first similarity For the second weight corresponding to the second similarity, the connectivity information, and the third weight corresponding to the connectivity information, the confidence of the target predicted trajectory is determined through an operation function; wherein, the first weight, the third weight The second weight, the third weight, and the operating function are preconfigured or derived from machine learning.

Through this design, when determining the confidence of the target predicted trajectory, weights can be configured for the first similarity, the second similarity and the connectivity information, so as to flexibly adjust the way of determining the confidence.

In a possible design, the confidence level of the predicted target trajectory is obtained according to a first similarity degree and a second similarity degree, wherein the first similarity degree is used to indicate that the historical trajectory of the target vehicle is different from the target vehicle's historical trajectory. the similarity between the first lanes in the target predicted trajectory, the second similarity is used to indicate the similarity between the predicted driving trajectory from the first position to the second position and the second lane Spend.

Through this design, the confidence of the target predicted trajectory is determined according to the first similarity between the historical trajectory of the target vehicle and the first lane in the target predicted trajectory, and the second similarity between the predicted driving trajectory and the second lane in the target predicted trajectory , the confidence of the target predicted trajectory obtained from this combines the similarity between the target predicted trajectory and the historical trajectory of the target vehicle and the two-stage trajectory of the predicted driving trajectory, so that the determined confidence of the target predicted trajectory can better reflect the target predicted trajectory. accuracy.

In a possible design, the first threshold is determined according to the following manner: acquiring a data set containing multiple trajectory prediction samples, and determining an end point position error value corresponding to the second position in each trajectory prediction sample of the data set; The first threshold is determined according to the confidence of each trajectory prediction sample, the end position error value corresponding to the second position in each trajectory prediction sample, and a preset value.

In a possible design, an average value of confidence levels of at least one target trajectory prediction sample is determined, and the determined average value is used as the first threshold, wherein the at least one target trajectory prediction sample is the multiple At least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is less than or equal to the preset value; Set the value of at least one trajectory prediction sample as a positive sample set, and use at least one trajectory prediction sample whose error value of the end position corresponding to the second position in the plurality of trajectory prediction samples is greater than the preset value as a negative sample set, based on the binary classification The algorithm determines a metric value for distinguishing the positive sample set and the negative sample set, and uses the determined metric value as the first threshold.

Through the above design, the first threshold can be determined according to the data set including multiple trajectory prediction samples, so that the target predicted trajectory obtained when candidate predicted trajectories are screened by the first threshold value is more suitable for the actual scene, and the accuracy of trajectory prediction is further improved.

In a second aspect, an embodiment of the present application provides a trajectory prediction method, the method includes:

determining a first set of lanes in the map according to the first position of the target vehicle; wherein the first position is the position of the target vehicle at a first moment, and the first set of lanes includes at least one first lane; Predict the second position of the target vehicle based on the set trajectory prediction model; wherein, the second position is the position of the target vehicle at a second moment, and the second moment is located after the first moment; A second set of lanes is determined in the map according to the second position; wherein, the second set of lanes includes at least one second lane; according to the first set of lanes and the second set of lanes, at least one set of lanes is determined A candidate predicted trajectory; wherein any candidate predicted trajectory includes a first lane in the first lane set and a second lane in the second lane set, and any candidate predicted trajectory includes a first lane Having connectivity with the second lane; determining the confidence of the at least one candidate predicted trajectory, and the confidence of any candidate predicted trajectory is used to indicate the similarity between the candidate predicted trajectory and the target trajectory expected to travel by the target vehicle ; Take the candidate predicted trajectory with the confidence greater than the first threshold as the target predicted trajectory.

Through the above method, when predicting the trajectory of the target vehicle, a first set of lanes can be determined in the map according to the first position of the target vehicle, and the first set of lanes includes lanes where the target vehicle may travel from the first position. Then, a second lane set is determined according to the second position of the target vehicle predicted by the set trajectory prediction model, and the second lane set includes lanes where the target vehicle may travel from the second position. Then determine at least one candidate predicted trajectory according to the first lane set and the second lane set, each candidate predicted trajectory includes a first lane in the first lane set and a second lane in the second lane set, and each candidate If the first lane and the second lane included in the predicted track are connected, the determined at least one candidate predicted track is determined according to the multi-stage position information of the target vehicle, and the accuracy is higher. Finally, the confidence of at least one candidate predicted trajectory is determined, and the candidate predicted trajectory whose confidence is greater than the first threshold is used as the target predicted trajectory. By selecting the target predicted trajectory, the candidate predicted trajectory that is more likely to be driven by the target vehicle can be selected, which further improves the accuracy of trajectory prediction.

In a possible design, determining at least one candidate predicted trajectory according to the first lane set and the second lane set includes: when the target second lane in the second lane set is different from the first lane set When there is connectivity in the first target lane of the target, a target candidate predicted trajectory including the target second lane and the target first lane is determined; wherein the target candidate predicted trajectory is included in the at least one candidate predicted trajectory .

In a possible design, determining the confidence level of the at least one candidate predicted trajectory includes:

The confidence level of the first candidate prediction track in the at least one candidate prediction track is determined by the following steps, wherein the first candidate prediction track is any one of the at least one candidate prediction track: obtaining the first candidate prediction track Connectivity information between the first lane and the second lane included in the predicted track; obtain the historical track of the target vehicle, and determine the first track between the first lane included in the first candidate predicted track and the historical track. a similarity; determining the predicted driving trajectory of the target vehicle from the first position to the second position based on the trajectory prediction model; determining the second lane included in the first candidate predicted trajectory and the predicted trajectory a second similarity between the driving trajectories; according to the first similarity, the second similarity, and the connectivity information, determine the confidence of the first candidate predicted trajectory.

In a possible design, determining the confidence of the first candidate predicted trajectory according to the first similarity, the second similarity, and the connectivity information includes:

The confidence level of the candidate predicted trajectory is determined according to the following formula:

P=f(w ₁ , P _h )*f(w ₂ , P _f )*f(w ₃ , EC)

Wherein, in the above formula, P is the confidence level of the first candidate predicted trajectory, w ₁ , w ₂ , and w ₃ are weight coefficients pre-configured or obtained according to machine learning, respectively; P _h is the first similarity; P _f is the second similarity; EC is the connectivity information, and f( ) is an operation function preset or obtained according to machine learning.

Through this design, when determining the confidence of the target predicted trajectory, weights can be configured for the first similarity, the second similarity and the connectivity information, and the confidence can be determined by a preset or an operation function obtained by machine learning, thereby The way of determining the confidence can be flexibly adjusted.

In a possible design, the determining the first set of lanes in the map according to the first position of the target vehicle includes: using at least one lane in the map that is within a preset query radius from the first position as the The first set of lanes; and the determining of a second set of lanes in the map according to the second position, comprising: using at least one lane in the map that is within a preset query radius from the second position as the second lane set.

Through this design, the first lane set and/or the second lane set can be determined according to the preset query radius, so as to determine the lane sets corresponding to the positions of the target vehicle at different times.

In a possible design, the first threshold is determined according to the following methods: acquiring a data set containing a plurality of trajectory prediction samples, and determining an end point position error value corresponding to the second position in each trajectory prediction sample in the data set; The confidence of each trajectory prediction sample, the end point position error value corresponding to the second position in each trajectory prediction sample, and a preset value determine the first threshold.

In a third aspect, an embodiment of the present application provides a trajectory prediction apparatus, including a unit for performing each step in any of the foregoing aspects.

In a fourth aspect, an embodiment of the present application provides a trajectory prediction apparatus, including a processor and a memory, where computer program instructions are stored in the memory, and when the trajectory prediction apparatus runs, the processor executes the method provided in any of the above aspects .

In a fifth aspect, the embodiments of the present application further provide a computer program, which, when the computer program runs on a computer, causes the computer to execute the method provided in any of the foregoing aspects.

In a sixth aspect, embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a computer, the computer is made to perform any of the above aspects provided method.

In a seventh aspect, an embodiment of the present application further provides a chip, where the chip is configured to read a computer program stored in a memory and execute the method provided in any one of the foregoing aspects.

In an eighth aspect, an embodiment of the present application further provides a chip system, where the chip system includes a processor for supporting a computer device to implement the method provided in any one of the foregoing aspects. In a possible design, the chip system further includes a memory for storing necessary programs and data of the computer device. The chip system can be composed of chips, and can also include chips and other discrete devices.

In a ninth aspect, an embodiment of the present application further provides a terminal, where the terminal may include the trajectory prediction apparatus described in the third aspect or the fourth aspect. Further, the terminal may be a vehicle.

Description of drawings

1 is a schematic structural diagram of a system to which a trajectory prediction method provided by an embodiment of the present application is applicable;

FIG. 2 is a schematic diagram of a lane provided by an embodiment of the present application;

3 is a flowchart of a trajectory prediction method provided by an embodiment of the present application;

Fig. 4 is a kind of map that the embodiment of this application provides;

FIG. 5 is a schematic diagram of a lane connectivity relationship provided by an embodiment of the present application;

FIG. 6 is a schematic diagram of a candidate prediction trajectory provided by an embodiment of the present application;

7 is a flowchart of a method for determining a confidence level of a first candidate prediction trajectory provided by an embodiment of the present application;

FIG. 8 is a schematic diagram of a method for determining a historical trajectory according to an embodiment of the present application;

9 is a schematic diagram of a predicted driving trajectory provided by an embodiment of the present application;

10 is a flowchart of a second trajectory prediction method provided by an embodiment of the present application;

11 is a schematic diagram of a map including multiple lanes provided by an embodiment of the present application;

12 is a schematic diagram of a first trajectory prediction method provided by an embodiment of the present application;

13 is a schematic diagram of a second trajectory prediction method provided by an embodiment of the present application;

14 is a schematic diagram of a third trajectory prediction method provided by an embodiment of the present application;

FIG. 15 is a schematic structural diagram of a first trajectory prediction apparatus provided by an embodiment of the application;

FIG. 16 is a schematic structural diagram of a second trajectory prediction apparatus provided by an embodiment of the present application.

Detailed ways

For the convenience of understanding the embodiments of the present application, terms related to the embodiments of the present application are introduced below:

Intelligent driving refers to the fact that machines help people to drive, and completely replace human driving under special circumstances. Specifically, relevant audio-visual signals and information are obtained through sensors on the vehicle, and the corresponding follow-up system is controlled through cognitive computing, so as to realize the analysis of the driving state of the vehicle and the control of the subsequent driving of the vehicle.

The embodiments of the present application will be further described below with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of the architecture of a system to which a trajectory prediction method provided by an embodiment of the present application is applicable. The system architecture includes at least two vehicles (as shown in FIG. 1 by taking n vehicles as an example, vehicle 1 and vehicle 1). 2. Vehicle n, where n is an integer greater than or equal to 2), and may also include a monitoring center. The monitoring center can receive the information or requests sent by the vehicle, monitor the vehicle, and control the driving of the vehicle by sending control instructions to the vehicle.

Referring to FIG. 1 , any vehicle may include: sensors, in-vehicle communication equipment 105 , high-precision positioning equipment 106 , vehicle controller 107 , and trajectory prediction device 108 . The sensor includes one or more of the following devices: long- and short-range millimeter-wave radar 101 , ultrasonic radar 102 , laser radar 103 , and vehicle-mounted camera 104 . Specifically, only the vehicle 1 is shown in FIG. 1 .

The functions of the above modules included in the vehicle are explained in detail below:

The long- and short-range millimeter wave radar 101 is a radar that works in the millimeter wave band (millimeter wave) for detection, and is used to collect the transmission time of light pulses reaching obstacles, and send the collected data to the trajectory prediction device 108; After the optical pulse transmission time is collected, data such as distance, speed, and azimuth angle of surrounding obstacles are calculated, and the calculated data is sent to the trajectory prediction device 108 .

Ultrasonic radar 102: It is a radar that detects the position of a target using ultrasonic waves. Its working principle is that the ultrasonic generator generates ultrasonic waves, and the probe of the ultrasonic radar 102 receives the ultrasonic waves reflected by the obstacles, and calculates the distance to the obstacles according to the time difference between the transmitted ultrasonic waves and the received reflected ultrasonic waves. The radar 102 may send the collected data to the trajectory prediction device 108 .

Lidar 103 : a radar system that emits a laser beam to detect characteristic quantities such as the position and velocity of a target. Its working principle is to transmit a detection signal (laser beam) to the target, and then compare the received signal (target echo) reflected from the target with the transmitted detection signal, and after proper processing, the relevant data of the target can be obtained. Such as target distance, bearing, altitude, speed, attitude, and even shape and other parameters. In this application, the lidar 103 is used to collect the signal reflected from the obstacle, and send the reflected signal and the transmitted signal to the trajectory prediction device 108; Comparing the signals, data such as distance and speed of surrounding obstacles are obtained by processing, and the data obtained by processing are sent to the trajectory prediction device 108 .

Vehicle-mounted camera 104: used to collect surrounding images or videos, and send the collected images or videos to the trajectory prediction device 108; wherein, the vehicle-mounted camera may be a monocular camera, a binocular camera, a depth camera, etc. No restrictions. In this application, the vehicle-mounted camera 104 can analyze the vehicle information such as the speed and distance of surrounding obstacles in the image or video after collecting the image or video, or analyze the lane information in the image or video, etc., and send the data obtained by the analysis. to the trajectory prediction device 108 .

In-vehicle communication device 105: a device used to communicate with other vehicles or monitoring centers, specifically, it can be used to receive vehicle information sent by other vehicles, such as the driving trajectories of other vehicles, or send its own trajectory to other surrounding vehicles, or communicate with the monitoring center. Interact, send vehicle information to the monitoring center or receive control commands sent by the monitoring center. For example, the in-vehicle communication device 105 may be a telematics BOX (TBOX).

High-precision positioning device 106: collect the precise position information of the current vehicle (with an error of less than 20cm), and the global positioning system (global positioning system, GPS) time information corresponding to the precise position information, and send the collected information to the track Prediction device 108 . The high-precision positioning device 108 may be a combined positioning system or a combined positioning module. The high-precision positioning device 108 may include a global navigation satellite system (GNSS), an inertial measurement unit (inertial measurement unit, IMU) and other devices and sensors. The global navigation satellite system can output global positioning information with a certain accuracy (eg, 5-10 Hz), and the frequency of the inertial measurement unit is generally high (eg, 1000 Hz). System information, output high-frequency precise positioning information (generally requires more than 200Hz).

Vehicle controller 107 : executes control commands to control vehicle steering, acceleration, deceleration, start, stop, and the like.

Trajectory prediction device 108: It can be set in the vehicle. The trajectory prediction device 108 is specifically implemented by a processor and a memory. The processor includes a central processing unit (CPU) or a device or module with processing functions. For example, the trajectory prediction device 108 may be an intelligent driving fusion perception module in the form of an intelligent driving domain controller, an in-vehicle electronic control unit (ECU), an in-vehicle mobile data center (Mobile Data Center, MDC), and the like. In this application, the trajectory prediction device 108 receives the target vehicle information sent by the sensor, such as target vehicle speed, target vehicle position information, etc., and the trajectory prediction device 108 obtains lane information, which can be specifically received through the in-vehicle communication device 105. Roadside units or Lane information sent by the monitoring center. Taking the trajectory prediction device of the first vehicle as an example, the trajectory prediction device 108 determines the target predicted trajectory of the target vehicle according to the target vehicle information and the lane information, and performs planning and control on the driving route of the first vehicle according to the target predicted trajectory, The control command is generated and sent to the vehicle controller 107, so that the vehicle controller 107 controls the first vehicle to execute the control command. Or, after receiving the vehicle information of the first vehicle sent by the sensor or the high-precision positioning device 106, the trajectory prediction device 108 sends the vehicle information of the first vehicle to the monitoring center through the in-vehicle communication device 105. Vehicle information and lane information of the area where the first vehicle is currently located, determine the predicted trajectory of the first vehicle, and send control commands to the first vehicle or other vehicles around the first vehicle according to the determined predicted trajectory, so that the monitoring center can monitor the traffic on the road. monitoring and control of vehicles.

It should be noted that the surrounding obstacles mentioned above refer to other vehicles here.

In the system shown in Fig. 1, taking the trajectory prediction of vehicle 1 to vehicle 2 as an example, vehicle 1 determines the current position (x, y) of vehicle 2, and takes the current position (x, y) of vehicle 2 as the origin, according to the prediction Set the query radius R to search for a lane set including at least one lane in the map, and for each lane in the found lane set, determine the vertical distance between the current position of the vehicle 2 and the lane, except that the vertical distance in the lane set is greater than the predetermined distance. After setting the thresholded lanes, a candidate lane set including at least one candidate lane is obtained. For example, the vehicle shown in FIG. 2 is vehicle 2 , and vehicle 1 determines that the candidate lanes corresponding to vehicle 2 are lane A and lane B as shown in FIG. 2 . For each candidate lane, determine the projection position of the current position of the target vehicle projected to the candidate lane, and determine the angle that the target vehicle needs to turn if it travels to the candidate lane according to the projection position and the median line of the candidate lane, and then determine the target vehicle's position. The heading angle is determined, and the angle difference between the heading angle and the included angle is determined. According to the order of the angle difference corresponding to the candidate lanes from small to large, the first N candidate lanes are selected to determine the predicted trajectory. Each predicted trajectory contains a candidate lane. Among them, N is a positive integer greater than or equal to 1. For example, if it is finally determined that the angle difference corresponding to lane A is smaller than the angle difference corresponding to lane B, the predicted trajectory of vehicle 2 can be determined according to lane A, such as the predicted trajectory shown in FIG. 2 .

It can be seen from the above that when a vehicle predicts the trajectory of the target vehicle, it only predicts the trajectory of the target vehicle based on the current position of the target vehicle, the vehicle heading angle and other information. At intersections, the projected lanes may be inaccurate and the predicted trajectory may be wrong; and when the driver has irregular driving behavior, the trajectory prediction is only based on the current position of the target vehicle and the vehicle heading angle information, which may also lead to The phenomenon that the predicted trajectory is unreasonable. Therefore, the current trajectory prediction methods are not accurate enough.

Based on the above problems, an embodiment of the present application provides a trajectory prediction method, which can be applied to any vehicle in the system shown in FIG. 1 , and can also be applied to a monitoring center in the system shown in FIG. 1 . When the trajectory prediction method provided by the present application is applied to any vehicle, taking the first vehicle as an example, the first vehicle can determine the target predicted trajectory of the target vehicle according to the trajectory prediction method, and the target vehicle can be a vehicle on the current road. Any other vehicle other than the first vehicle, or the first vehicle can also predict the target predicted trajectory of the vehicle according to the trajectory prediction method provided in this application, and the first vehicle can adjust the driving trajectory or speed of the vehicle according to the target predicted trajectory. , to achieve intelligent driving. When the trajectory prediction method provided by this application is applied to the monitoring center, the monitoring center can predict the trajectory of the target vehicle. At this time, the target vehicle can be any vehicle on the current road. The monitoring center determines the target predicted trajectory of the target vehicle. The target predicted trajectory of the target vehicle sends control commands to the target vehicle and the surrounding vehicles of the target vehicle, and flexibly controls and monitors the vehicles driving on the road.

The trajectory prediction method provided by the embodiment of the present application is further introduced by taking the application of the trajectory prediction method to the first vehicle as an example. The first vehicle may be any vehicle in the system shown in FIG. 1 . FIG. 3 is a flowchart of a trajectory prediction method provided by an embodiment of the present application, and the trajectory prediction method includes the following steps:

S301: Acquire a first position of the target vehicle at a first moment, and determine a first lane set corresponding to the first position.

Optionally, the first moment can be the moment when the first vehicle triggers the trajectory prediction of the target vehicle, or the preset specified moment when the first vehicle performs trajectory prediction on the target vehicle, or can be the system time when S301 is executed ( referred to as the current moment).

In an optional implementation manner, the first vehicle obtains the first position of the target vehicle through a sensor, or the first vehicle receives the first position sent by the target vehicle through the in-vehicle communication device, and the first position sent by the target vehicle may be the target vehicle. Obtained through a device such as a high-precision positioning device in the target vehicle, and this application does not specifically limit the obtaining method.

After determining the first position of the target vehicle, the first vehicle acquires first lane information, where the first lane information includes position information of all or part of the lanes in the area to which the first position belongs. Optionally, the first lane information may be preconfigured in the first vehicle, or the first vehicle may acquire the first lane information through vehicle to everything (V2X) communication, for example, the first vehicle communicates with the roadside unit through an in-vehicle communication device. Or the monitoring center communicates to obtain the first lane information; or the first vehicle obtains the first lane information through sensors in the first vehicle, for example, the first vehicle detects and perceives the first lane information in real time through sensors such as on-board cameras and lidars.

The first vehicle determines a first set of lanes according to the first location and the first lane information.

In an optional embodiment, the first vehicle obtains a map of the area to which the first position belongs, obtains first lane information according to the map, and obtains at least a distance from the first position within a preset query radius according to the first lane information. One lane is set as the first lane. For example, referring to the map shown in FIG. 4 , obtain the first lane information according to the map, and take the first position of the target vehicle as the origin to determine the lane within the preset query radius, such as lane A, If both lane B and lane C are within the preset query radius, it is determined that the first lane set includes three first lanes, lane A, lane B, and lane C.

S302: Predict the second position of the target vehicle at the second moment, and determine the second lane set corresponding to the second position.

Wherein, the second time is located after the first time in time, for example, the second time is a time t seconds later than the first time.

In an optional embodiment, the second position of the target vehicle is predicted by a set trajectory prediction model, wherein the set trajectory prediction model can be a constant velocity motion model (Constant Velocity Model, CV), a constant velocity constant rotation rate model ( Constant Turn Rate and Velocity, CTRV), Long Short-Term Memory Neural Network (Long Short-Term Memory, LSTM). Optionally, the running speed and acceleration of the target vehicle are obtained, specifically, the running speed and acceleration of the target vehicle may be determined by a sensor in the first vehicle, and the first position, running speed and acceleration of the target vehicle are used as the set trajectory prediction model. The input feature of , obtains the second position of the output of the set trajectory prediction model.

It should be noted that, when the first vehicle in the embodiment of the present application determines the second position of the target vehicle, a trajectory prediction model can be selected as the set trajectory prediction model, and the set trajectory prediction model can predict at least one first position of the target vehicle. Two positions, or multiple trajectory prediction models may be selected to predict multiple second positions of the target vehicle, which is not limited in this application.

After determining the second position of the target vehicle, the first vehicle determines a second set of lanes according to the second lane information, where the second lane information includes position information of all or part of the lanes in the area to which the second position belongs.

An optional implementation is that the first vehicle obtains a map of the area to which the second position belongs, obtains second lane information according to the map, and obtains at least a distance from the second position within a preset query radius according to the second lane information. One lane is set as the second lane.

For the specific manner in which the first vehicle determines the second lane set, reference may be made to the manner in which the first vehicle determines the first lane set in S301 , and the repetition will not be repeated. It should be noted that the first lane set and the second lane set may be the same or different. For example, the first position and the second position belong to the same area, the acquired first lane set and the second lane set are the same, or the first position Different from the area to which the second position belongs, the acquired first lane set and the second lane set are different.

S303: Determine at least one candidate predicted trajectory according to the first lane set and the second lane set.

Optionally, the candidate predicted track includes a first lane in the first lane set and a second lane in the second lane set, and the first lane and the second lane included in the candidate predicted track have connectivity. The connectivity between the first lane and the second lane means that the first lane and the second lane are connected, and it can be understood that when the target vehicle is driving in the first lane, it can pass to the second lane. The connectivity between lanes is a lane attribute, which can be obtained through lane information. For example, after the first vehicle obtains a map of the area to which the first position belongs, lane information including lane connectivity information is obtained from the map. When there is connectivity between lanes, the connectivity relationship between lanes can be direct connectivity, lane change connectivity, ramp connectivity, straight lane change, etc. For example, Figure 5 shows several common lane connectivity relationships. The

numbers

1, 2, 3, 4, 5, 6, and 7 in Figure 5 correspond to a lane, respectively. The directly connected lanes are: 1 and lane 3, lane 3 and lane 6, lane 2 and lane 4, lane 4 and lane 7; the lanes with lane change connections are: lane 6 and lane 7; the lanes with ramp connections are lane 4 and lane 5; there are The lanes for straight lane change are lane 1 and lane 2, lane 3 and lane 4.

In an optional implementation manner, the connectivity information between lanes can be represented by a connectivity coefficient between lanes, for example, the connectivity coefficient between two directly connected lanes is 1, and the two lanes connected by changing lanes The connectivity coefficient between is 0.8 etc.

In a specific implementation, the first vehicle determines at least one candidate predicted trajectory according to the following methods:

The first vehicle determines, for each first lane in the first lane set, a second lane that has connectivity with the first lane in the second lane set according to the connectivity between the lanes. The first vehicle determines at least one candidate predicted trajectory from each of the first lanes and the second lane with connectivity to the first lane. For example, in the map shown in Figure 6, the first set of lanes includes lane D, lane E, and the second set of lanes includes lane F, lane G, and lane H; for lane D, there is connectivity with lane D The second lane is lane G and lane H, then the candidate predicted trajectories determined according to lane D and lane G are shown in Figure 6 as track 1, and the candidate predicted trajectories determined according to lane D and lane H are shown in Figure 6 track 2. For the lane E, the second lane that has connectivity with the lane E is the lane F, then the candidate predicted trajectory determined according to the lane E and the lane F is the trajectory 3 shown in FIG. 6 .

S304: Determine the confidence level of the at least one candidate predicted trajectory.

Optionally, the confidence level of any candidate predicted trajectory is used to indicate the similarity between the candidate predicted trajectory and the target trajectory expected to travel by the target vehicle, wherein the target trajectory expected to travel by the target vehicle is the driver expected to drive the target vehicle. The trajectory of the target vehicle or the running trajectory of the target vehicle within a preset time period in the future determined by the trajectory prediction device in the target vehicle.

In an optional embodiment, the first vehicle determines a confidence level for each candidate predicted trajectory in the at least one candidate predicted trajectory. Taking the first candidate predicted trajectory in the at least one candidate predicted trajectory as an example, the first candidate predicted trajectory is: Any one of the at least one candidate prediction trajectory, FIG. 7 is a flowchart of a method for determining the confidence of the first candidate prediction trajectory, the method comprising the following steps:

S701: Acquire connectivity information between a first lane and a second lane included in the first candidate predicted trajectory.

In an optional implementation manner, connectivity information between lanes may be preconfigured in the first vehicle, or the first vehicle may acquire connectivity information between lanes through V2X communication. The roadside unit or the monitoring center obtains the connectivity information between the lanes; or, the first vehicle obtains the connectivity information between the lanes through a sensor in the first vehicle, for example, the first vehicle is detected in real time by sensors such as an on-board camera and lidar Connectivity between lanes to obtain connectivity information between lanes. Optionally, the lane information may also include connectivity information between lanes.

S702: Determine the first similarity between the first lane included in the first candidate predicted track and the historical track.

The historical trajectory is the trajectory that the target vehicle has traveled within the preset historical period. Optionally, the historical trajectory of the target vehicle can be obtained according to the following methods:

Method 1. The first vehicle obtains N frames of historical position information of the target vehicle within a historical preset time period. Optionally, the first vehicle obtains N frames of historical position information of the target vehicle within a historical preset time period through a sensor or a vehicle-mounted camera. information, the vehicle fits the acquired historical position information of the target vehicle N frames to obtain the historical trajectory, for example, Figure 8 is a schematic diagram of a historical trajectory determination method, the rectangle in Figure 8 is the acquired target vehicle within the historical time period The N frame position information of , and the historical trajectory shown in Figure 8 is obtained by fitting the N frame position information.

Mode 2: The first vehicle receives the historical track sent by the target vehicle through V2X communication.

After the historical trajectory of the target vehicle is obtained, the first lane can be determined by calculating any one of the Euclidean distance, cosine similarity, Pearson parameter or Tanimoto similarity between the first lane in the first candidate predicted trajectory and the historical trajectory. A similarity, for example, when the first similarity is determined by calculating the Euclidean distance between the first lane and the historical track, the first similarity can be determined according to the following formula 1:

When the first similarity is determined by calculating the cosine similarity between the first lane and the historical trajectory, the first similarity can be determined according to the following formula 2:

When the first similarity is determined by calculating the Pearson parameter between the first lane and the historical track, the first similarity can be determined according to the following formula 3:

When the first similarity is determined by calculating the Tanimoto similarity between the first lane and the historical track, the first similarity can be determined according to the following formula 4:

Among them, in the above formula, P _h is the first similarity between the first lane and the historical track, _xi is the ith sampling point after sampling the median line of the first lane, and y _i is the historical track. The i-th sampling point after sampling, n is the number of sampling points.

For example, assuming that the first lane in the first candidate predicted trajectory is the lane I shown in FIG. 8 , and the determined historical trajectory is the historical trajectory shown in FIG. 8 , the Euclidean distance between the lane I and the historical trajectory is calculated. , and determine the first similarity according to the Euclidean distance.

S703: Determine the second similarity between the second lane included in the first candidate predicted trajectory and the predicted driving trajectory.

The predicted travel trajectory is the predicted travel trajectory of the target vehicle from the first position to the second position.

In an optional implementation manner, the first vehicle determines the predicted travel trajectory through a set trajectory prediction model. The set trajectory prediction model can be a constant velocity model (Constant Velocity Model, CV), a constant velocity constant rate model (Constant Turn Rate and Velocity, CTRV), and a long short-term memory neural network (Long Short-Term Memory, LSTM). Optionally, the running speed and acceleration of the target vehicle are obtained, specifically, the running speed and acceleration of the target vehicle may be determined by a sensor in the first vehicle, and the first position, running speed and acceleration of the target vehicle are used as the set trajectory prediction model. to obtain the predicted driving trajectory output by the set trajectory prediction model.

It should be noted that when the first vehicle in the embodiment of the present application determines the predicted driving trajectory of the target vehicle, a trajectory prediction model can be selected as the set trajectory prediction model, and the set trajectory prediction model can predict at least one prediction of the target vehicle. The driving trajectory, or a plurality of trajectory prediction models may be selected to predict the plurality of predicted driving trajectories of the target vehicle, which is not limited in this application. In addition, in the embodiment of the present application, when the first vehicle determines the second position of the target vehicle through the set trajectory prediction model, the predicted driving trajectory of the target vehicle may be determined through the same or different trajectory prediction models. When the first vehicle determines the predicted driving trajectory When the trajectory prediction model of the trajectory is different from the trajectory prediction model for determining the second position, the second position corresponding to the predicted driving trajectory may be used as the input of the trajectory prediction model when determining the predicted driving trajectory.

Optionally, the second similarity may be determined by calculating any one of Euclidean distance, cosine similarity or Pearson parameter between the second lane in the first candidate predicted trajectory and the predicted driving trajectory. For a specific calculation method, reference may be made to the implementation of calculating the first similarity in S702, and repeated details will not be repeated.

For example, assuming that the second lane included in the first candidate predicted trajectory is lane J as shown in Figure 9, the predicted driving trajectory obtained based on CTRV is the trajectory F1 shown in Figure 9, and the predicted driving trajectory obtained based on LSTM is For the track F2 shown in FIG. 9, the second similarity between the lane J and the track F1 and the second similarity between the lane J and the track F2 are calculated respectively, and the second similarity between the lane J and the track F1 is selected. The second similarity with a larger numerical value among the similarity and the second similarity between the lane J and the trajectory F2 is used as the second similarity between the lane J and the predicted travel trajectory.

It should be noted that the method for calculating the similarity in the embodiment of the present application is not limited to the above-mentioned method for obtaining the similarity by calculating the Euclidean distance, the cosine similarity or the Pearson parameter. Any method for calculating the similarity between two trajectories or line segments. methods are applicable.

S704: According to the first similarity between the first lane in the first candidate predicted track and the historical track, the second similarity between the second lane in the first candidate predicted track and the predicted driving track, and the first lane and the The connectivity information between the second lanes determines the confidence of the first candidate predicted trajectory.

Optionally, the confidence level of the first candidate predicted trajectory can be determined according to the following formula 5:

P=f(w ₁ , P _h )*f(w ₂ , P _f )*f(w ₃ , EC) Equation 5

Among them, in the above formula, P is the confidence level of the first candidate predicted trajectory, w ₁ , w ₂ , and w ₃ are weight coefficients pre-configured or obtained according to machine learning, respectively; P _h is the first lane in the first candidate predicted trajectory The first similarity with the historical trajectory; P _f is the second similarity between the second lane in the first candidate predicted trajectory and the predicted driving trajectory; EC is the connectivity between the first lane and the second lane information, f() is an operation function that is preset or obtained according to machine learning.

In a possible implementation manner, the operation function in the above formula 4 can be a function preset by the technician, such as multiplication, then the confidence level of the first candidate predicted trajectory can be determined according to formula 6:

P=(w ₁ *P _h )*(w ₂ *P _f )*(w ₃ *EC) Equation 5

In another possible implementation manner, the operation function in the above formula 4 may be a pooling operation in a convolutional neural network, and only data larger than a preset value is calculated.

In yet another possible implementation, the operation function in the above formula 4 may be a machine learning model. Specifically, the P _f , P _h and EC corresponding to the first candidate prediction trajectory are input into the confidence calculation model, and the confidence is obtained. The degree of confidence of the first candidate predicted trajectory output by the model is calculated. Optionally, the confidence calculation model can be trained in the following ways:

The initial confidence calculation model is trained according to the data set. The data set contains multiple training samples, wherein each training sample includes the predicted trajectory, the actual driving trajectory, the first and second lanes included in the predicted trajectory, and the first and second lanes corresponding to the predicted trajectory. At least one feature among a similarity, a second similarity and connectivity information, a predicted similarity between the actual driving track and the predicted track, and the like. The first similarity, second similarity and connectivity information corresponding to the predicted trajectory in each training sample are used as the input of the initial confidence calculation model, the initial confidence calculation model is trained, and the confidence of the predicted trajectory output by the model is calculated. and the loss value between the predicted similarity, and adjust the model parameters according to the loss value obtained after each round of training until the loss value converges in the preset range.

It should be noted that, since the first lane and the second lane included in the candidate predicted trajectory in the embodiment of the present application have connectivity, when calculating the confidence level of the candidate predicted trajectory, the first lane in the candidate predicted trajectory can also be calculated only according to the The confidence is determined by the first similarity with the historical trajectory and the second similarity between the second lane in the candidate predicted trajectory and the predicted driving trajectory, so as to simplify the calculation of the confidence and improve the trajectory prediction efficiency.

S305: Use the candidate predicted trajectory with the confidence greater than the first threshold as the target predicted trajectory.

Wherein, the first threshold may be an empirical value preset by the technician, or the first threshold may be determined according to the following methods:

Based on the trajectory prediction method provided by the embodiment of the present application, a data set including multiple trajectory prediction samples is obtained, and each trajectory prediction sample includes a historical trajectory, a second position and a predicted driving trajectory, a candidate predicted trajectory and its confidence level, and the actual driving of the vehicle. track and other data; determine the end position error value corresponding to the second position in each trajectory prediction sample in the data set; specifically, determine the actual position of the vehicle at the second moment according to the actual driving trajectory of the vehicle, and determine the first position according to the second position and the actual position. The error value of the end position corresponding to the two positions. The first threshold is determined according to the confidence of each trajectory prediction sample, the end point position error value corresponding to the second position in each trajectory prediction sample, and a preset value.

In an optional embodiment, at least one target trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples of the data set is less than or equal to a preset value is determined, and at least one target trajectory prediction sample is determined. The average value of the confidence level of , and the determined average value is used as the first threshold.

In another optional implementation, at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples of the data set is less than or equal to a preset value is used as a positive sample set, and the data set At least one trajectory prediction sample whose end position error value corresponding to the second position is greater than the preset value among the plurality of trajectory prediction samples is regarded as a negative sample set, and is determined based on a binary classification algorithm for distinguishing the positive sample set and the negative sample set. The metric value is determined, and the determined metric value is used as the first threshold. The binary classification algorithm may be a classification algorithm or a clustering algorithm.

It should be noted that there can be multiple preset values, which correspond to different application scenarios or functional modules. For example, the preset value corresponding to the high-speed driving scene is 5 meters, and the preset value corresponding to the daily driving scene is 2 meters. Different preset values may determine first thresholds corresponding to different application scenarios or functional modules.

Through the above method, when the first vehicle predicts the trajectory of the target vehicle, the first lane set can be determined in the map according to the first position of the target vehicle, and the first lane set includes lanes where the target vehicle may travel from the first position. Then, a second lane set is determined according to the second position of the target vehicle predicted by the set trajectory prediction model, and the second lane set includes lanes where the target vehicle may travel from the second position. The first vehicle then determines at least one candidate predicted trajectory according to the first lane set and the second lane set, each candidate predicted trajectory includes a first lane in the first lane set and a second lane in the second lane set, and The first lane and the second lane included in each candidate predicted trajectory are connected, and the determined at least one candidate predicted trajectory is determined according to the multi-stage position information of the target vehicle, and the accuracy is higher. Finally, the first vehicle determines the confidence of at least one candidate predicted trajectory, and uses the candidate predicted trajectory whose confidence is greater than the first threshold as the target predicted trajectory, and uses the confidence to represent the similarity between the candidate predicted trajectory and the target trajectory that the target vehicle expects to travel, The target predicted trajectory is selected according to the confidence, and the candidate predicted trajectory that is more likely to be driven by the target vehicle can be selected, which further improves the accuracy of trajectory prediction.

10 is a flowchart of another trajectory prediction method provided by an embodiment of the present application, and the trajectory prediction method includes the following steps:

S1001: Acquire a first position of the target vehicle at a first moment, and determine a first lane set corresponding to the first position.

S1002: Predict the second position of the target vehicle at the second moment, and determine a second lane set corresponding to the second position.

It should be noted that, for the specific implementation of S1001 to S1002, reference may be made to S301 to S302 of FIG. 3 in the embodiment of the present application, and repeated details will not be repeated.

S1003: Determine at least a pair of lane combinations according to the first lane set and the second lane set.

The at least one pair of lane combinations includes a first lane in the first lane set and a second lane in the second lane set.

S1004: Determine the confidence of at least one pair of lane combinations.

In the specific implementation, reference may be made to the method of determining the confidence level of the candidate predicted trajectory shown in FIG. 7 . It should be noted that since the first lane and the second lane in the lane combination determined in S1004 in this embodiment may not exist Connectivity, for the first and second lanes that do not have connectivity, the connectivity coefficient between the first and second lanes can be set to 0, or the first and second lanes are opposite lanes to each other When , the connectivity coefficient between the first lane and the second lane can be set to -1 to further reduce the confidence of the lane combination that does not have a connectivity relationship.

S1005: Determine the target predicted trajectory according to the lane combination with the confidence greater than the first threshold.

In a specific implementation, a lane combination with a confidence greater than the first threshold may be used as the target predicted trajectory.

Through the above method, the confidence can be determined for all lane combinations in the first lane set and the second lane set, without the need to filter the lanes that can form candidate predicted trajectories according to the lane connectivity relationship, but by setting the absence of connectivity relationship. The connectivity coefficient between the lanes is set to 0 or a negative value, and when selecting a lane combination with a confidence greater than the first threshold, the lane combination that does not have a connectivity relationship is filtered, and then the target prediction trajectory is determined to improve the trajectory prediction. s efficiency.

The confidence level of the candidate predicted trajectory calculated in the trajectory prediction method provided in the embodiment of the present application can be used to select the target predicted trajectory, and can also be used to select the target of interest. For example, after the confidence of at least one candidate predicted lane is determined for the target vehicle, if the proportion of the confidence of at least one candidate predicted trajectory of the target vehicle that is lower than the preset value is greater than the preset ratio, it means that it is difficult to predict the driving of the target vehicle track, select the target vehicle as the object of interest. If it is believed that the target vehicle has a problem of ambiguous driving behavior and may violate traffic rules, it is necessary to pay attention to the target vehicle.

It should be noted that, for the convenience of description, the above description of the trajectory prediction method provided by the embodiment of the present application takes the application of the trajectory prediction method to the first vehicle as an example. It can be understood that the trajectory prediction method provided by the embodiment of the present application can also be It is applied to one or more functional modules in the vehicle, or, the trajectory prediction method provided in the embodiment of the present application can also be applied to the monitoring center in the system shown in FIG. 1 . For specific implementation, see the trajectory prediction method shown in FIG. , and the repetition will not be repeated.

The trajectory prediction method provided by the embodiment of the present application is further introduced below with several examples:

Figure 11 shows a schematic diagram of a map containing multiple lanes, and the map includes lanes 1-6. Assuming that the connectivity coefficient between lanes is used as the connectivity information between lanes, the connectivity between some lanes The coefficients are shown in the table below:

车道Lane	连通性系数 Connectivity Coefficient

车道1和车道2Lane 1 and Lane 2	11
车道1和车道4 Lane 1 and Lane 4	0.80.8
车道4和车道5 Lane 4 and Lane 5	0.80.8
车道4和车道6 Lane 4 and Lane 6	0.60.6

Fig. 12 is a schematic diagram of the first kind of trajectory prediction method when the target vehicle travels on the lane included in the map shown in Fig. 11 , Fig. 12 shows the historical trajectory of the target vehicle and the predicted travel based on the set trajectory prediction model Track, the first vehicle determines that the first lane set includes lane 1, where the first similarity between lane 1 and track 1 is 0.7, the second lane set includes lane 2 and lane 4, and the second similarity between lane 2 and track 2 is 0.5, and the second similarity between lane 4 and track 2 is 0.8.

The first vehicle determines at least one candidate predicted trajectory according to the connected first lane and the second lane, respectively candidate predicted trajectory 1 (including lane 1 and lane 2) and candidate predicted trajectory 2 (including lane 1 and lane 4).

1. Calculate the confidence of the candidate predicted trajectory 1:

P1=0.7*0.5*1=0.35

2. Calculate the confidence of the candidate predicted trajectory 2:

P1=0.7*0.8*0.8=0.448

Assuming that the first threshold value is 0.3, the candidate predicted trajectory 1 and the candidate predicted trajectory 2 with confidence greater than the first threshold are determined as the target predicted trajectory.

FIG. 13 is a schematic diagram of the trajectory prediction method when the second target vehicle is driving on the lane included in the map shown in FIG. 10 . FIG. 13 shows the historical trajectory of the target vehicle and the trajectory obtained based on the set trajectory prediction model. F1, track F2 and track F3. The first vehicle determines that the first set of lanes includes lane 1, where the first similarity between lane 1 and the historical track is 0.9, the second set of lanes includes lane 2, the second similarity between lane 2 and track F1 is 0.8, and the second similarity between lane 2 and the track F1 is 0.8. The second similarity of track F2 is 0.9, and the second similarity between lane 2 and track F3 is 0.7.

The first vehicle determines a candidate predicted trajectory 1 (including lane 1 and lane 2) according to the first and second lanes with connectivity, and determines the confidence level of the candidate predicted trajectory 1: P1=0.9*max(0.8,0.9, 0.7)*1=0.81.

Assuming that the first threshold is 0.8, the confidence of the candidate predicted track 1 is greater than the first threshold, and the candidate predicted track 1 is used as the target predicted track.

FIG. 14 is a schematic diagram of the driving trajectory of the third target vehicle when driving on the lane included in the map shown in FIG. 10 . FIG. 14 shows the historical trajectory of the target vehicle and the predicted driving based on the set trajectory prediction model. Track, the first vehicle determines that the first set of lanes includes lane 4, where the first similarity between lane 4 and the historical track is 0.4, the second set of lanes includes lane 5 and lane 6, and the second similarity between lane 5 and the predicted driving trajectory is is 0.8, and the second similarity between lane 6 and the predicted driving trajectory is 0.6.

The first vehicle determines at least one candidate predicted trajectory according to the connected first lane and the second lane, respectively candidate predicted trajectory 1 (including lane 4 and lane 5) and candidate predicted trajectory 2 (including lane 4 and lane 6).

1. Calculate the confidence of the candidate predicted trajectory 1:

P1=0.4*0.6*0.8=0.192

2. Calculate the confidence of the candidate predicted trajectory 2:

P1=0.4*0.9*0.6=0.216

Assuming that the first threshold is 0.2, the candidate predicted trajectory 2 whose confidence is greater than the first threshold is determined to be the target predicted trajectory.

Based on the same technical concept, the present application also provides a trajectory prediction device 1500, the trajectory prediction device 1500 can be applied to any vehicle or monitoring center in the system shown in FIG. 1, and FIG. 15 is the trajectory prediction device A schematic structural diagram of the apparatus 1500 , the trajectory prediction apparatus 1500 includes an acquisition unit 1501 and a processing unit 1502 . The functions of each unit in the trajectory prediction apparatus 1500 will be introduced below.

The obtaining unit 1501 is configured to obtain a first position of the target vehicle at a first moment, where the first position corresponds to a first lane set including at least one first lane;

The processing unit 1502 is configured to determine a target predicted trajectory, the confidence of the target predicted trajectory is greater than or equal to a first threshold, the target predicted trajectory is at least one of at least one candidate predicted trajectory, the at least one candidate predicted trajectory Any candidate predicted trajectory in the trajectory includes a first lane in the first lane set and a second lane in the second lane set, and the first lane and the second lane are connected in the candidate predicted trajectory;

Wherein, the second lane set corresponds to a second position, the second lane set includes at least one second lane, and the second position is the predicted position of the target vehicle at a second moment, the second moment After the first instant in time.

In one embodiment, the first set of lanes is determined according to the first position and first lane information; and/or the second set of lanes is determined according to the second position and second lane information The first lane information includes the position information of all or part of the lanes in the area to which the first position belongs, and the second lane information includes the position information of all or part of the lanes in the area to which the second position belongs. , the first lane information and the second lane information are the same or different.

In one embodiment, the processing unit 1502 is further configured to:

Obtain the first lane information according to the map of the area to which the first position belongs, and obtain the second lane information according to the map of the area to which the second position belongs; At least one lane is used as the first set of lanes, and/or at least one lane within a preset query radius from the second position is used as the second set of lanes.

In one embodiment, the processing unit 1502 is further configured to:

Before determining the target predicted trajectory, the at least one candidate predicted trajectory is determined according to the first lane set and the second lane set.

In one embodiment, the processing unit 1502 is specifically configured to: when there is connectivity between the target second lane in the second lane set and the target first lane in the first lane set, determine whether to include the target second lane in the first lane set. target candidate predicted trajectories of the target second lane and the target first lane; wherein the target candidate predicted trajectory is included in the at least one candidate predicted trajectory.

In one embodiment, the confidence of the target predicted trajectory is obtained according to a first similarity, a second similarity and connectivity information, where the first similarity is used to indicate the history of the target vehicle The similarity between the trajectory and the first lane in the target predicted trajectory, and the second similarity is used to indicate the difference between the predicted driving trajectory from the first position to the second position and the second lane. The connectivity information is used to indicate the connectivity between the first lane and the second lane.

In an implementation manner, the processing unit 1502 is specifically configured to determine the confidence of the target predicted trajectory according to the following manner: according to the first similarity, the first weight corresponding to the first similarity, the The second similarity, the second weight corresponding to the second similarity, the connectivity information, and the third weight corresponding to the connectivity information are used to determine the confidence of the target predicted trajectory through an operation function; The first weight, the second weight, the third weight and the operation function are pre-configured or obtained according to machine learning.

In an implementation manner, the processing unit 1502 is specifically configured to determine the first threshold according to the following manner:

Obtain a data set containing multiple trajectory prediction samples, and determine the end position error value corresponding to the second position in each trajectory prediction sample of the data set; The first threshold is determined by predicting the end position error value corresponding to the second position in the sample and the preset value.

In an embodiment, the processing unit 1502 is specifically configured to: determine the average value of the confidence levels of at least one target trajectory prediction sample, and use the determined average value as the first threshold, wherein the at least one One target trajectory prediction sample is at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is less than or equal to a preset value; or the second position in the plurality of trajectory prediction samples corresponding to At least one trajectory prediction sample whose end position error value is less than or equal to a preset value is used as a positive sample set, and at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is greater than the preset value is used. The sample is taken as a set of negative samples, and a metric value for distinguishing the set of positive samples from the set of negative samples is determined based on a binary classification algorithm, and the determined metric value is used as the first threshold.

Based on the same technical concept, the present application also provides a trajectory prediction apparatus 1600. FIG. 16 is a schematic structural diagram of a trajectory prediction apparatus 1600 provided by an embodiment of the present application. The trajectory prediction apparatus 1600 may be applied to the apparatus shown in FIG. 1 . any vehicle or monitoring center in the system. Referring to FIG. 16 , the trajectory prediction apparatus 1600 includes: a processor 1601 , a memory 1602 and a bus 1603 . Among them, the processor 1601 and the memory 1602 communicate through the bus 1603, and the communication can also be realized through other means such as wireless transmission. The memory 1602 is used for storing instructions, and the processor 1601 is used for executing the instructions stored in the memory 1602 . The memory 1602 stores program codes, and the processor 1601 can call the program codes stored in the memory 1602 to perform the following operations:

obtaining a first position of the target vehicle at a first moment, the first position corresponding to a first lane set including at least one first lane; determining a target predicted trajectory, the confidence of the target predicted trajectory is greater than or equal to a first threshold, The target predicted trajectory is at least one of at least one candidate predicted trajectory, and any candidate predicted trajectory in the at least one candidate predicted trajectory includes the first lane in the first lane set and the first lane in the second lane set. Two lanes, the first lane and the second lane are connected in the candidate predicted trajectory; wherein the second lane set corresponds to a second position, the second lane set includes at least one second lane, the The second position is the predicted position of the target vehicle at a second time, which is located after the first time in time.

In one embodiment, the processor 1601 is further configured to:

Obtain the first lane information according to the map of the area to which the first position belongs, and obtain the second lane information according to the map of the area to which the second position belongs; At least one first lane is used as the first set of lanes, and/or at least one lane within a preset query radius from the second position is used as the second set of lanes.

In one embodiment, the processor 1601 is further configured to:

In an implementation manner, the processor 1601 is specifically configured to: when there is connectivity between the target second lane in the second lane set and the target first lane in the first lane set, determine whether to include the target candidate predicted trajectories of the target second lane and the target first lane; wherein the target candidate predicted trajectory is included in the at least one candidate predicted trajectory.

In an implementation manner, the processor 1601 is specifically configured to determine the confidence of the target predicted trajectory according to the following manner: according to the first similarity, the first weight corresponding to the first similarity, the The second similarity, the second weight corresponding to the second similarity, the connectivity information, and the third weight corresponding to the connectivity information are used to determine the confidence of the target predicted trajectory through an operation function; The first weight, the second weight, the third weight and the operation function are pre-configured or obtained according to machine learning.

In an implementation manner, the processor 1601 is specifically configured to determine the first threshold according to the following manner:

In an embodiment, the processor 1601 is specifically configured to: determine an average value of the confidence level of at least one target trajectory prediction sample, and use the determined average value as the first threshold, wherein the at least one One target trajectory prediction sample is at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is less than or equal to a preset value; or the second position in the plurality of trajectory prediction samples corresponding to At least one trajectory prediction sample whose end position error value is less than or equal to a preset value is used as a positive sample set, and at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is greater than the preset value is used. The sample is taken as a set of negative samples, and a metric value for distinguishing the set of positive samples from the set of negative samples is determined based on a binary classification algorithm, and the determined metric value is used as the first threshold.

It will be appreciated that the memory 1602 in FIG. 16 of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Wherein, the non-volatile memory may be a read-only memory (Read-Only Memory, ROM), a programmable read-only memory (Programmable ROM, PROM), an erasable programmable read-only memory (Erasable PROM, EPROM), an electrically programmable read-only memory (Erasable PROM, EPROM). Erase programmable read-only memory (Electrically EPROM, EEPROM) or flash memory. Volatile memory may be Random Access Memory (RAM), which acts as an external cache. By way of illustration and not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synchlink DRAM, SLDRAM) ) and direct memory bus random access memory (Direct Rambus RAM, DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but not be limited to, these and any other suitable types of memory.

Based on the above embodiments, the embodiments of the present application further provide a computer program, when the computer program runs on a computer, the computer causes the computer to execute the trajectory prediction method provided by the embodiment shown in FIG. 3 or FIG. 4 .

Based on the above embodiments, the embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a computer, the computer executes the program shown in FIG. 3 or FIG. 4 . The trajectory prediction method provided by the illustrated embodiment. The storage medium may be any available medium that the computer can access. By way of example and not limitation, computer readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or be capable of carrying or storing instructions or data structures in the form of desired program code and any other medium that can be accessed by a computer.

Based on the above embodiments, an embodiment of the present application further provides a chip for reading a computer program stored in a memory to implement the trajectory prediction method provided by the embodiment shown in FIG. 3 or FIG. 4 .

Based on the above embodiments, an embodiment of the present application provides a chip system, where the chip system includes a processor for supporting a computer device to implement the trajectory prediction method shown in FIG. 3 or FIG. 4 . In a possible design, the chip system further includes a memory for storing necessary programs and data of the computer device. The chip system may be composed of chips, or may include chips and other discrete devices.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to the present application. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the protection scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

A trajectory prediction method, characterized in that the method comprises:

acquiring a first position of the target vehicle at a first moment, where the first position corresponds to a first lane set including at least one first lane;

Determine a target predicted trajectory, the confidence level of the target predicted trajectory is greater than or equal to the first threshold, the target predicted trajectory is at least one of at least one candidate predicted trajectory, and any candidate predicted trajectory in the at least one candidate predicted trajectory Including the first lane in the first lane set and the second lane in the second lane set, the first lane and the second lane are connected in the candidate predicted trajectory;

Wherein, the second lane set corresponds to a second position, the second lane set includes at least one second lane, and the second position is the predicted position of the target vehicle at a second moment, the second moment After the first instant in time.
The method of claim 1, wherein the first set of lanes is determined according to the first location and first lane information; and/or

the second set of lanes is determined according to the second position and second lane information;

Wherein, the first lane information includes the position information of all or part of the lanes in the area to which the first position belongs, the second lane information includes the position information of all or part of the lanes in the area to which the second position belongs, and the The first lane information and the second lane information are the same or different.
The method of claim 2, wherein the method further comprises:

Obtain the first lane information according to the map of the area to which the first position belongs, and obtain the second lane information according to the map of the area to which the second position belongs;

Taking at least one lane within a preset query radius from the first position as the first set of lanes, and/or taking at least one lane within a preset query radius from the second position as the first lane set Two-lane collection.
The method according to any one of claims 1 to 3, characterized in that before determining the predicted target trajectory, the method further comprises:

The at least one candidate predicted trajectory is determined from the first set of lanes and the second set of lanes.
The method of claim 4, wherein the determining the at least one candidate predicted trajectory according to the first lane set and the second lane set comprises:

When there is connectivity between the target second lane in the second lane set and the target first lane in the first lane set, determine a target candidate predicted trajectory including the target second lane and the target first lane ; wherein the target candidate prediction track is included in the at least one candidate prediction track.
The method according to any one of claims 1 to 5, wherein the confidence of the target predicted trajectory is obtained according to a first similarity, a second similarity and connectivity information, wherein the first The similarity is used to indicate the similarity between the historical trajectory of the target vehicle and the first lane in the target predicted trajectory, and the second similarity is used to indicate the distance from the first position to the second position The similarity between the predicted driving trajectory and the second lane, and the connectivity information is used to indicate the connectivity between the first lane and the second lane.
The method of claim 6, wherein the confidence of the target predicted trajectory is determined according to the following methods:

According to the first similarity, the first weight corresponding to the first similarity, the second similarity, the second weight corresponding to the second similarity, the connectivity information, and the connectivity information The corresponding third weight is used to determine the confidence of the target predicted trajectory through the operation function;

Wherein, the first weight, the second weight, the third weight and the operation function are pre-configured or obtained according to machine learning.
The method according to any one of claims 1 to 7, wherein the first threshold is determined according to the following manner:

Acquire a data set containing multiple trajectory prediction samples, and determine the end position error value corresponding to the second position in each trajectory prediction sample of the data set;

The first threshold is determined according to the confidence of each trajectory prediction sample, the end position error value corresponding to the second position in each trajectory prediction sample, and a preset value.
The method of claim 8, wherein the method further comprises:

Determine the average value of the confidence of at least one target trajectory prediction sample, and use the determined average value as the first threshold, wherein the at least one target trajectory prediction sample is the second of the plurality of trajectory prediction samples At least one trajectory prediction sample whose end point position error value corresponding to the position is less than or equal to the preset value; or

Take at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is less than or equal to the preset value as the positive sample set, and use the trajectory prediction sample corresponding to the second position in the plurality of trajectory prediction samples. At least one trajectory prediction sample whose end position error value is greater than a preset value is used as a negative sample set, and a metric value for distinguishing the positive sample set and the negative sample set is determined based on a binary classification algorithm, and the determined metric value is as the first threshold.
A trajectory prediction device, comprising:

an obtaining unit, configured to obtain a first position of the target vehicle at a first moment, where the first position corresponds to a first lane set including at least one first lane;

A processing unit, configured to determine a target predicted trajectory, the confidence level of the target predicted trajectory is greater than or equal to a first threshold, the target predicted trajectory is at least one of at least one candidate predicted trajectory, and the at least one candidate predicted trajectory Any candidate predicted trajectory includes a first lane in the first lane set and a second lane in the second lane set, and the first lane and the second lane are connected in the candidate predicted trajectory;

Wherein, the second lane set corresponds to a second position, the second lane set includes at least one second lane, and the second position is the predicted position of the target vehicle at a second moment, the second moment After the first instant in time.
The apparatus of claim 10, wherein the first set of lanes is determined according to the first location and first lane information; and/or

the second set of lanes is determined according to the second position and second lane information;

Wherein, the first lane information includes the position information of all or part of the lanes in the area to which the first position belongs, the second lane information includes the position information of all or part of the lanes in the area to which the second position belongs, and the The first lane information and the second lane information are the same or different.
The apparatus of claim 11, wherein the processing unit is further configured to:

Obtain the first lane information according to the map of the area to which the first position belongs, and obtain the second lane information according to the map of the area to which the second position belongs;

Taking at least one lane within a preset query radius from the first position as the first set of lanes, and/or taking at least one lane within a preset query radius from the second position as the first lane set Two-lane collection.
The device according to any one of claims 10 to 12, wherein the processing unit is further configured to:

Before determining the target predicted trajectory, the at least one candidate predicted trajectory is determined according to the first lane set and the second lane set.
The apparatus of claim 13, wherein the processing unit is specifically configured to:

When there is connectivity between the target second lane in the second lane set and the target first lane in the first lane set, determine a target candidate predicted trajectory including the target second lane and the target first lane ; wherein the target candidate prediction track is included in the at least one candidate prediction track.
The apparatus according to claim 14, wherein the confidence of the target predicted trajectory is obtained according to a first similarity, a second similarity and connectivity information, wherein the first similarity is used to indicate The similarity between the historical trajectory of the target vehicle and the first lane in the target predicted trajectory, and the second similarity is used to indicate that the predicted driving trajectory from the first position to the second position is the same as the predicted driving trajectory. the similarity between the second lanes, and the connectivity information is used to indicate the connectivity between the first lane and the second lane.
The apparatus according to claim 15, wherein the processing unit is specifically configured to determine the confidence level of the predicted target trajectory according to the following manner:

According to the first similarity, the first weight corresponding to the first similarity, the second similarity, the second weight corresponding to the second similarity, the connectivity information, and the connectivity information The corresponding third weight is used to determine the confidence of the target predicted trajectory through the operation function;

Wherein, the first weight, the second weight, the third weight and the operation function are pre-configured or obtained according to machine learning.
The apparatus according to any one of claims 10 to 16, wherein the processing unit is specifically configured to determine the first threshold according to the following manner:

Acquire a data set containing multiple trajectory prediction samples, and determine the end point position error value corresponding to the second position in each trajectory prediction sample of the data set;

The first threshold is determined according to the confidence of each trajectory prediction sample, the end position error value corresponding to the second position in each trajectory prediction sample, and a preset value.
The apparatus of claim 17, wherein the processing unit is specifically configured to:

Determine the average value of the confidence of at least one target trajectory prediction sample, and use the determined average value as the first threshold, wherein the at least one target trajectory prediction sample is the second of the plurality of trajectory prediction samples At least one trajectory prediction sample whose end point position error value corresponding to the position is less than or equal to the preset value; or

Take at least one trajectory prediction sample whose end position error value corresponding to the second position in the plurality of trajectory prediction samples is less than or equal to the preset value as a positive sample set, and use the trajectory prediction sample corresponding to the second position in the plurality of trajectory prediction samples. At least one trajectory prediction sample whose end position error value is greater than a preset value is used as a negative sample set, and a metric value for distinguishing the positive sample set and the negative sample set is determined based on a binary classification algorithm, and the determined metric value is as the first threshold.
A trajectory prediction apparatus, characterized in that it includes a processor and a memory, wherein computer program instructions are stored in the memory, and when the trajectory prediction apparatus runs, the processor executes the computer program instructions stored in the memory to The operational steps of the method of any one of the preceding claims 1 to 9 are implemented.
A computer-readable storage medium, characterized by comprising computer instructions, which, when executed by a processor, cause a reader/writer to perform the method according to any one of claims 1 to 9.
A computer program product, characterized in that, when the computer program product runs on a processor, the in-vehicle device is made to execute the method according to any one of claims 1 to 9.