CN110196429A - Vehicle target recognition methods, storage medium, processor and system - Google Patents

Abstract

The invention discloses a vehicle target recognition method, a storage medium, a processor and a system. Wherein, the method includes: obtaining point cloud data of the surrounding environment from the vehicle-mounted laser radar; clustering the point cloud data to obtain multiple categories; merging the multiple categories; using a singular value decomposition algorithm to analyze the point cloud data performing straight line fitting; and performing vehicle target recognition according to the straight line fitting result.

Description

Vehicle target recognition method, storage medium, processor and system

technical field

The present invention relates to the field of automobile automatic driving, in particular, to a vehicle target recognition method, storage medium, processor and system based on vehicle laser radar.

Background technique

In order to reduce the occurrence of traffic accidents and improve vehicle safety, the research and development of automatic driving systems that assist the driver or replace the driver to control the vehicle is becoming a hot topic in the field of vehicle engineering. Vehicle object recognition is an important function and the basis for its reliable operation in automotive automatic driving systems. By detecting and identifying other vehicles on the road, the automatic driving system can obtain dynamic driving environment information in a timely manner, thereby providing a basis for driving processes such as path planning, obstacle avoidance and collision avoidance.

In existing automatic driving systems, vehicle target recognition usually relies on visual sensors or ultrasonic radar sensors. The visual sensor method uses photosensitive elements to simulate the visual system of the human eye, shoots the driving environment, and obtains the dynamic image information of the current environment, and then uses image processing technology to simulate the processing of the image by the human brain, recognizes the target from the image and obtains the target's information. Motion parameters, so as to realize the perception of real space information. This method has simple equipment, low cost, and can achieve a long working distance, but it is extremely sensitive to changes in the lighting conditions, weather conditions, and traffic conditions of the road environment, and has poor environmental adaptability. The ultrasonic radar sensor method is to install ultrasonic radar on the vehicle, use the reflection of ultrasonic waves to detect the vehicle target, and measure the speed of the vehicle target according to the time difference between ultrasonic emission and return. This method has good environmental adaptability and can work stably under changing environmental conditions. At the same time, because the wave velocity of ultrasonic waves in the air is relatively slow, the structural information along the propagation direction contained in the echo signal is easy to detect. It has high resolving power, so its accuracy is relatively high. However, the ultrasonic frequency is low, the wavelength is long, and the distance that can work reliably is relatively short. Usually, the effective detection distance is only within 5 meters.

In recent years, with the large-scale application of lidar in the vehicle field, a new technical solution has been provided for the automotive automatic driving system. Lidar is rapidly changing the field of autonomous driving with its high detection accuracy, fast detection speed, strong anti-interference ability, and good environmental adaptability. However, how to process the rich detection data of lidar to obtain applicable information with good real-time performance is still an important problem to be solved in the application of vehicle lidar.

Aiming at the problem that the visual sensor method to identify vehicle targets is extremely sensitive to changes in road environment lighting conditions, weather conditions, and traffic conditions, and the environmental adaptability is relatively poor, and the ultrasonic radar sensor method can reliably work at a relatively short distance, no effective solution has yet been proposed.

Contents of the invention

Embodiments of the present invention provide a vehicle target recognition method, storage medium, processor and system, to at least solve the problem that the visual sensor method for vehicle target recognition is extremely sensitive to changes in the illumination conditions, weather conditions, and traffic conditions of the road environment, and the environmental adaptability comparison Poor, the ultrasonic radar sensor method to identify the vehicle target can work reliably at a relatively short distance.

According to an aspect of an embodiment of the present invention, a vehicle target recognition method is provided, including:

Obtain the point cloud data of the surrounding environment from the vehicle lidar; cluster the point cloud data to obtain multiple categories; merge multiple categories; use the singular value decomposition algorithm to fit the point cloud data to a straight line; and according to the straight line The fitting results are used for vehicle target recognition.

According to another aspect of the embodiments of the present invention, there is also provided a storage medium, the storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the vehicle object recognition method described in any one of the above.

According to another aspect of the embodiments of the present invention, a processor is also provided, and the processor is used to run a program, wherein the vehicle target recognition method described in any one of the above is executed when the program is running.

According to another aspect of the embodiments of the present invention, there is also provided a vehicle target recognition system, including: a processor; The point cloud data of the surrounding environment; clustering the point cloud data to obtain multiple categories; merging multiple categories; using the singular value decomposition algorithm to perform straight line fitting on the point cloud data; Target Recognition.

In the embodiment of the present invention, by using a two-dimensional laser radar as a sensor for vehicle target recognition, the shortcomings of the visual sensor being greatly affected by the environment and the short detection distance of the ultrasonic radar sensor are overcome, and the environmental perception ability of the automatic driving vehicle is improved. The safety and adaptability of driving vehicles has a positive meaning.

Description of drawings

The accompanying drawings described here are used to provide a further understanding of the present invention and constitute a part of the application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention. In the attached picture:

FIG. 1 is a schematic diagram of a vehicle-mounted terminal executing a vehicle target recognition method according to an embodiment of the present invention;

Fig. 2 is an architecture diagram of a vehicle-mounted two-dimensional laser radar point cloud data processing module according to an embodiment of the present invention;

FIG. 3 is a flow chart of a vehicle target recognition method according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of removing breakpoints in the method according to an embodiment of the present invention;

5A and 5B are schematic diagrams of the results of straight line fitting in the method according to the embodiment of the present invention;

6 is a schematic diagram of target recognition according to the result of straight line fitting in the method according to an embodiment of the present invention; and

FIG. 7 is a flowchart of a detailed description of the method according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a vehicle object recognition system according to an aspect of an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only It is an embodiment of a part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

It should be noted that the terms "first" and "second" in the description and claims of the present invention and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the invention described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

According to an embodiment of the present invention, a method embodiment of a vehicle object recognition is provided. It should be noted that the steps shown in the flow chart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and, Although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that shown or described herein.

The method embodiment provided in Embodiment 1 of the present application may be executed in a mobile terminal, a vehicle-mounted terminal, or a similar computing device. Fig. 1 shows a block diagram of the hardware structure of a vehicle-mounted terminal (or mobile device) for realizing the vehicle target recognition method. As shown in Figure 1, the vehicle-mounted terminal 10 (or mobile device 10) may include one or more (shown by 102a, 102b, ..., 102n in the figure) processor 102 (the processor 102 may include but not limited to a microprocessor A processing device such as a processor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, it can also include: a display, an input/output interface (I/O interface), a universal serial bus (USB) port (which can be included as one of the ports of the I/O interface), a network interface, a power supply and/or camera. Those of ordinary skill in the art can understand that the structure shown in FIG. 1 is only a schematic diagram, and it does not limit the structure of the above-mentioned electronic device. For example, the vehicle terminal 10 may also include more or fewer components than those shown in FIG. 1 , or have a different configuration than that shown in FIG. 1 .

It should be noted that the one or more processors 102 and/or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuit may be implemented in whole or in part as software, hardware, firmware or other arbitrary combinations. In addition, the data processing circuit can be a single independent processing module, or be fully or partially integrated into any one of the other components in the vehicle terminal 10 (or mobile device). As mentioned in the embodiment of the present application, the data processing circuit is used as a processor control (for example, the selection of the terminal path of the variable resistor connected to the interface).

The memory 104 can be used to store software programs and modules of application software, such as the program instruction/data storage device corresponding to the vehicle target recognition method in the embodiment of the present invention, and the processor 102 runs the software programs and modules stored in the memory 104, thereby Executing various functional applications and data processing, that is, realizing the above-mentioned vulnerability detection method of the application program. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory that is remotely located relative to the processor 102 , and these remote memories may be connected to the vehicle terminal 10 through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission device 106 is used to receive or transmit data via a network. The specific example of the above network may include a wireless network provided by the communication provider of the vehicle terminal 10 . In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet in a wireless manner.

The display can be, for example, a touch-screen liquid crystal display (LCD), which enables the user to interact with the user interface of the vehicle terminal 10 (or mobile device).

Under the above operating environment, the present application provides a vehicle identification method as shown in FIG. 3 . Fig. 3 is a flowchart of a vehicle identification method according to an embodiment of the present invention.

As shown in Figure 3, an embodiment of the present invention provides a method for detecting obstacles, including:

S302: Obtain point cloud data of the surrounding environment from the vehicle lidar;

S304: Perform clustering processing on the point cloud data to obtain multiple categories;

S306: Merging multiple categories;

S308: Using a singular value decomposition algorithm to fit a straight line to the point cloud data; and

S310: Perform vehicle target recognition according to the straight line fitting result.

Specifically, a vehicle target recognition method based on vehicle-mounted two-dimensional laser radar data provided by the present invention mainly uses a clustering method to distinguish the point cloud data generated by the two-dimensional laser radar, and then distinguishes the well-discriminated point cloud The data is fitted with a straight line (for example, the SVD decomposition method can be used for straight line fitting), and the vehicle target is recognized according to the straight line fitting result to complete the recognition of the vehicle target. At the same time, the vehicle target recognition method provided by the present invention includes a target tracking method, which can continuously track and identify the target.

The vehicle target recognition method, storage medium, processor and system provided by the embodiments of the present invention solve the problem that the visual sensor method for recognizing vehicle targets is extremely sensitive to changes in the illumination conditions, weather conditions, and traffic conditions of the road environment, and has poor environmental adaptability. The radar sensor method recognizes the technical problem of the relatively short distance at which vehicle targets can reliably work.

Specifically, Fig. 2 shows a structure diagram of a vehicle-mounted two-dimensional lidar point cloud data processing module that implements the method of the present invention. The vehicle-mounted two-dimensional laser radar point cloud data processing method provided by the present invention mainly includes four parts: a data preprocessing module 1 , a feature extraction module 2 , a target recognition module 3 and a target tracking module 4 . Among them, after the data preprocessing module 1 organizes the format of the original point cloud data, the point cloud data with two-dimensional coordinate information and speed information is divided into multiple different categories by applying a clustering algorithm, and the separated points are divided into different categories in the module. The categories are divided according to the region of interest, and the class categories in the region of interest are extracted and combined appropriately to reduce the amount of calculation. After the preprocessing module, the point cloud data is sent to the feature extraction module 2. In this module, after proper clustering and merging, the data will be screened for breakpoints. After removing outliers, the point cloud data will be decomposed by SVD. The method performs straight line fitting to obtain the vehicle shape profile of I-shape or L-shape (that is, straight line or broken line). After the outline is input into the target recognition module, it will be supplemented into a complete rectangle, and then the shape and size of the rectangle will be compared with those of different models, and combined with the speed information of the point for screening, so as to complete the vehicle target recognition. Finally, the identified vehicle target will be sent to the target tracking module 4, which predicts the position and state of the identified vehicle target at the next time according to the coordinate information of the point obtained at the previous time through the time update equation of the Kalman filter, And continue to search and update, so as to maintain the tracking and identification of the target.

Optionally, the category of the vehicle target recognition method in the present invention is determined by setting the geometric distance between two adjacent points. The specific clustering rule adopted here is: when the geometric distance between two adjacent points is less than a certain threshold, the two adjacent points are classified into the same category. Let the distance information of any two points output by the lidar be r _k and r _k+1 respectively, r _min = min{r _k, r _k+1 }, r _k,k+1 =|r _k+1 -r _k |, then the clustering rules can be expressed as follows:

In the formula, β is a parameter introduced to reduce the dependence of the segmented part on the distance from the lidar to the object, which can be determined through experiments. C ₀ is a parameter used to adjust the longitudinal error of the lidar. If C ₀ =0, then β represents the maximum absolute inclination angle of the target when two adjacent points belong to the same category. If β is too small, it is easy to classify points on the same object into different categories, and if β is too large, points on different objects may be classified into the same category. Usually, according to the angle and speed of the vehicle running on the actual road, the values of β and _C0 can be taken under different working conditions according to Table 1.

Table 1 Reference table for the values of β and C ₀ under different working conditions

Optionally, the clustering process of the vehicle target recognition method in the present invention includes dividing the scattered two-dimensional point cloud data into multiple point classes, wherein the geometric coordinates of the points within the point classes are similar.

Optionally, the operation of merging multiple categories includes: selecting an interest area and merging the categories in the area. In practical applications, since the width of the lane is limited, it is only necessary to detect the range within the lane. Selecting an appropriate detection range is beneficial to reduce the calculation amount of the vehicle target recognition method of the present invention and improve the real-time performance of the method. At the same time, for each category output by the clustering algorithm, it should be considered that the points on the same vehicle are divided into multiple categories due to occlusion or reflectance changes, and the categories should be properly merged to reduce the calculation amount of the method. In terms of the specific rules for the division of interest areas, since the width of the standard road lane in my country is 3.75m, and most standard roads have three lanes, the approximate width of the three lanes is 12m, that is, the left and right 6m of the laser radar installation position are horizontal interest areas , the vertical interest area is taken as 50m according to the actual measured effective working distance of the white target by the vehicle-mounted two-dimensional lidar. At the same time, when the lidar limit detection distance of 100m is exceeded, even if there is a target, the laser reflectivity is too low to be detected. Therefore, the area more than 100m away from the lidar is regarded as a non-target area. Based on the above, the screening and judging conditions for the target of interest can be obtained: x represents the horizontal distance from the point to the laser radar, and y represents the vertical distance from the point to the laser radar. If the interest _label = 0, it means that the target is out of range and is an invalid target. No processing; if interest _label = 1, it means that it is an object of interest, and it will continue to be processed in the next process; if interest _label = 2, it means that it is an object of no interest within the range, and it will not be processed.

After filtering out the categories in the region of interest, use the following function to merge the categories:

segment(f,l,θ _f, θ _l ,interest _label, d _max )

In this function, f is the starting point data of the category, l is the ending point data of the category, θ _f is the angle of the starting point, θ _l is the angle of the ending point, interest _label is the label of the category, and d _max is the laser distance in the category The distance from the farthest point of the radar to the lidar is compared according to the data obtained by each category. When the error percentage of these data is below the threshold, the categories can be merged. In practical applications, the threshold percentage can be an average of 10 %.

Optionally, the operation of selecting the region of interest includes: setting the measurement angle and distance according to the application scenario, and selecting the region of interest.

Optionally, the vehicle target recognition method in the present invention merges the categories according to the appearance and contour features of the vehicle target and the reason why the vehicle is divided into multiple categories.

Optionally, the straight line fitting operation of the vehicle target recognition method in the present invention includes: removing breakpoints from the clustered point cloud data to remove outliers. Specifically, the preprocessed point cloud data is firstly eliminated according to the principle shown in Figure 4 to remove obvious outliers. The specific method is: connect the starting point and the end point of the category into a straight line, if the distance between the points in the category and the straight line exceeds the set threshold, the

Points are breakpoints, and then further refine the search for breakpoints between the starting point and the breakpoint, the breakpoint and the end point, and iterate continuously, so that the collection of point cloud data is getting better and better. Due to the large size of the vehicle, the above-mentioned breakpoint search only removes breakpoints for objects whose number of points is greater than 5, and the distance between the start and end points of the category is greater than 0.5m and less than 12.5m. The threshold is set to 0.35m. The point cloud data after discontinuity elimination adopts the SVD decomposition method for straight line fitting. The specific operation is: for the category lattice A∈(m,n) formed by two-dimensional points, first calculate A ^T A and A ^T , Then find out the eigenvalues λ ₁ and λ ₂ of the two, and the eigenvector and Substitute the eigenvalues and eigenvectors into the formula Find the singular value σ _i , and finally get the singular value decomposition of the category lattice A A=U∑V ^T (VV ^T ＝E,UU ^T ＝E,∑＝diag[σ _1, σ ₂ …σ _i ,0,0 …0]), when using this method to fit a straight line to point cloud data, two types of straight lines as shown in Figure 5A can be obtained, among which type I fitting (that is, straight line), the equation of the straight line is ax+by+ c=0, when L-type fitting (that is, broken line), set the linear equation as ax+by+c=0 and bx-ay+d=0, the two can write the following two linear equations respectively, for linear Using the above method to solve the singular value decomposition of the equation system, the fitted straight line equation can be obtained.

Optionally, the straight line fitting operation includes: obtaining a straight line or a broken line through straight line fitting. Referring to FIG. 5A and FIG. 5B , the above-mentioned straight line (such as the I-shaped straight line V1 shown in FIG. 5A ) or broken line (such as the L-shaped straight line V2 shown in FIG. 5B ) can be obtained by straight line fitting.

Optional operations for vehicle target recognition include: complementing the straight line or folded line into a complete rectangle; comparing the size of the rectangle with the vehicle target, and screening based on the speed information of the points to complete the recognition of the vehicle target. Specifically, as shown in Figure 6, the straight line fitting result is input into the vehicle target recognition module. In this module, the I-type straight line obtained by fitting is used as the broad side, and the distance from the farthest point of interest to the broad side is used as the long side to complement Rectangle, or take the longer segment of the L-shaped straight line obtained by fitting as the broadside, and take the distance from the farthest point of interest to the broadside as the long side to complement the rectangle, and obtain the rectangular projection of the vehicle outline as shown in Figure 6. According to the length and width of the vehicle target, the vehicle target is divided into three types: small vehicle, medium vehicle and large vehicle. The standard of small vehicle is: length <3.6m, width <1.4m, standard size 3.6m×1.4m; The standard for medium-sized vehicles is: 3.6m<length<4.5m, 1.5m<width<2m, the standard size is 4.5m×1.8m; the standard for large vehicles is: 4.5m<length<12m, 2m<width<2.5m, standard The size is 8m×2.4m. At the same time, in order to deal with the situation where the target is occluded, it should also be judged in conjunction with the speed information of the point. When the speed information of the point is > 20km/h, it should be judged as a vehicle. After the above object recognition process, vehicle objects are recognized and classified into large, medium and small vehicles.

Optionally, the terminal vehicle target identification method of the present invention further includes: using a Kalman filter tracker to continuously track the target. Specifically, according to the Kalman filter model, the priori state quantity and the covariance of the error between the priori state quantity and the estimated quantity at the previous moment are calculated for the data of the target point obtained at a certain moment, according to the priori state quantity and The covariance of the error estimates the position of the target point at the next moment, and searches around the estimated position. When the new lidar measurement data is obtained, the new lidar measurement data is used to update the posterior state quantity and the posteriori The error covariance between the state quantity and the estimator can realize the continuous tracking of the target. Through continuous tracking and detection, the accuracy of the system is improved.

The method described in the embodiment of the present application will be described in detail below with reference to FIG. 7 .

Step S702: Install the vehicle-mounted 2D laser radar and connect it. Use the threaded mounting holes on the 2D LiDAR to fix the 2D LiDAR to the threaded holes of the number plate hanging on the front of the car with screws. After the two-dimensional laser radar is installed, connect the power line and signal line of the two-dimensional laser radar respectively. The specific connection method is: insert the USB power cable of the two-dimensional laser radar into the 12V power plug of the car cigarette lighter, Plug the RJ45 signal output cable into the communication network port of the on-board computer. Before the system enters the working state, the lidar parameters should be set in the following table 1. The purpose of this setting is to offset part of the measurement error caused by the movement of the lidar position in the working state by setting, and at the same time reduce the lidar due to The measurement error caused by pixel mixing and overlapping when the laser is emitted.

Table 1. Vehicle-mounted two-dimensional lidar parameter setting table under different working conditions

Step S704: Perform clustering processing on the point cloud data. After starting the vehicle and starting to drive, the two-dimensional lidar will collect a series of points containing two-dimensional coordinate information and speed information, and use the clustering algorithm to distinguish these point cloud data, and the obtained data cluster is a collection of a group of data objects , these objects are similar to each other from objects in the same cluster, and different from objects in other clusters. According to the specific application scenario of the vehicle-mounted two-dimensional lidar, the specific clustering rule adopted here is: when the geometric distance between two adjacent points is less than a certain set threshold, the two adjacent points are classified into the same category Inside. Let the distance information of any two points output by the lidar be r _k and r _k+1 respectively, r _min = min{r _k ,r _k+1 }, r _k,k+1 =|r _k+1 -r _k |, then the clustering rules can be expressed as follows:

In the formula, β is a parameter introduced to reduce the dependence of the segmented part on the distance from the lidar to the object, which can be determined through experiments. C ₀ is a parameter used to adjust the longitudinal error of the lidar. If C ₀ =0, then β represents the maximum absolute inclination angle of the target when two adjacent points belong to the same category. If β is too small, it is easy to classify points on the same object into different categories, and if β is too large, points on different objects may be classified into the same category. Usually, according to the angle and speed of the vehicle running on the actual road, the values of β and _C0 can be taken under different working conditions as shown in Table 2.

Table 2 Reference table for the values of β and C ₀ under different working conditions

Step S706: Select the interest area and merge the categories in the area. In practical applications, since the width of the lane is limited, it is only necessary to detect the range within the lane. Selecting an appropriate detection range is beneficial to reduce the calculation amount of the vehicle target recognition method of the present invention and improve the real-time performance of the method. At the same time, for each category output by the clustering algorithm, it should be considered that the points on the same vehicle are divided into multiple categories due to occlusion or reflectance changes, and the categories should be properly merged to reduce the calculation amount of the method. In terms of the specific rules for the division of interest areas, since the width of the standard road lane in my country is 3.75m, and most standard roads have three lanes, the approximate width of the three lanes is 12m, that is, the left and right sides of the laser radar installation position are each 6m for the horizontal interest area , the vertical interest area is taken as 50m according to the actual measured effective working distance of the white target by the vehicle-mounted two-dimensional lidar. At the same time, when the lidar limit detection distance of 100m is exceeded, even if there is a target, the laser reflectivity is too low to be detected. Therefore, the area more than 100m away from the lidar is regarded as a non-target area. Based on the above, the screening and judging conditions for the target of interest can be obtained: x represents the horizontal distance from the point to the laser radar, and y represents the vertical distance from the point to the laser radar. If the interest _label = 0, it means that the target is out of range and is an invalid target. No processing; if interest _label = 1, it means that it is an object of interest, and it will continue to be processed in the next process; if interest _label = 2, it means that it is an object of no interest within the range, and it will not be processed.

After filtering out the categories in the region of interest, use the following function to merge the categories:

segment(f,l,θ _f ,θ _l ,interest _label ,d _max )

In this function, f is the starting point data of the category, l is the ending point data of the category, θ _f is the angle of the starting point, θ _l is the angle of the ending point, interest _label is the label of the category, and d _max is the laser distance in the category The distance from the farthest point of the radar to the lidar is compared according to the data obtained by each category. When the error percentage of these data is below the threshold, the categories can be merged. In practical applications, the threshold percentage can be an average of 10 %.

Step S708: use SVD decomposition to perform straight line fitting on the point cloud data. The preprocessed point cloud data is firstly removed according to the principle shown in Figure 4 to remove obvious outliers. The specific method is: connect the starting point and the end point of the category into a straight line, if the distance from the point in the category to the line exceeds the set threshold, then the point is a breakpoint, and then separate the starting point and the breakpoint, the The search for the breakpoint is further refined between the breakpoint and the end point, and continuous iteration is carried out, so that the collection of point cloud data is getting better and better. Due to the large size of the vehicle, the above-mentioned breakpoint search only removes breakpoints for objects whose number of points is greater than 5, and the distance between the start and end points of the category is greater than 0.5m and less than 12.5m. The threshold is set to 0.35m. The point cloud data after breakpoint elimination adopts the SVD decomposition method for straight line fitting. The specific operation is: for the clustering lattice A∈(m,n) formed by two-dimensional points, first calculate A ^T A and A A ^T , and then find their eigenvalues λ ₁ and λ ₂ , and the eigenvector and Substitute the eigenvalues and eigenvectors into the formula Find the singular value σ _i , and finally get the singular value decomposition of the clustering lattice A A=U∑V ^T (VV ^T ＝E,UU ^T ＝E,∑＝diag[σ ₁ ,σ ₂ …σ _i ,0, 0…0]), when using this method to fit the point cloud data to a straight line, two types of straight lines can be obtained as shown in Figure 5A and Figure 5B, among them, when fitting type I (that is, straight line V1), the equation of the straight line is set ax+by+c=0, L-type fitting (that is, broken line V2), set the linear equation as ax+by+c=0 and bx-ay+d=0, both can write the following two A system of linear equations, using the above method to solve the singular value decomposition of the system of linear equations, can get the fitted line equation.

Step S710: Perform vehicle target recognition according to the straight line fitting result. Input the straight line fitting results shown in Figure 5A and Figure 5B into the vehicle target recognition module. In this module, the fitted I-type straight line is used as the broad side, and the distance from the farthest point of interest to the broad side is used as the long side to supplement Rectangle, or take the longer segment of the L-shaped straight line obtained by fitting as the broad side, and take the distance from the farthest point of interest to the broad side as the long side to complement the rectangle, and obtain the rectangular projection of the vehicle outline as shown in Figure 6. According to the length and width of the vehicle target, the vehicle target is divided into three types: small vehicle, medium vehicle and large vehicle. The standard of small vehicle is: length <3.6m, width <1.4m, standard size 3.6m×1.4m; The standard for medium-sized vehicles is: 3.6m<length<4.5m, 1.5m<width<2m, the standard size is 4.5m×1.8m; the standard for large vehicles is: 4.5m<length<12m, 2m<width<2.5m, standard The size is 8m×2.4m. At the same time, in order to deal with the situation where the target is occluded, it should also be judged in conjunction with the speed information of the point. When the speed information of the point is > 20km/h, it should be judged as a vehicle. After the above object recognition process, vehicle objects are recognized and classified into large, medium and small vehicles.

Step S712: Use the Kalman filter tracker to continuously track the target. According to the Kalman filter model, the priori state quantity and the covariance of the error of the prior state quantity and the estimated quantity at the previous moment are calculated for the data of the target point obtained at a certain moment, and the covariance of the prior state quantity and the error is calculated. The variance estimates the position of the target point at the next moment, and searches around the estimated position. When the new lidar measurement data is obtained, the new lidar measurement data is used to update the posterior state quantity and the posterior state quantity and The error covariance of the estimator enables continuous tracking of the target.

In summary, the vehicle target recognition method, storage medium, processor, and system provided by the embodiments of the present invention solve the problem that the visual sensor method for identifying vehicle targets is extremely sensitive to changes in the illumination conditions, weather conditions, and traffic conditions of the road environment, and the comparison of environmental adaptability. Poor, the ultrasonic radar sensor method to identify the vehicle target can work reliably at a relatively short distance.

In addition, referring to FIG. 1 , according to the second aspect of this embodiment, a storage medium 104 is also provided. The storage medium 104 includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the vehicle object recognition method described in any one of the above.

In addition, referring to FIG. 1 , according to a third aspect of this embodiment, a processor is also provided, and the processor is configured to run a program. Wherein, the vehicle target recognition method described in any one of the above items is executed when the program is running.

In addition, referring to FIG. 8 , according to the fourth aspect of this embodiment, a vehicle object recognition system 800 is also provided. The system 800 includes: a processor 802; and a memory 804, which is connected to the processor 802 and is used to provide the processor 802 with instructions for processing the following processing steps: acquiring point cloud data of the surrounding environment from the vehicle laser radar; Cluster processing to obtain multiple categories; merge multiple categories; use the singular value decomposition algorithm to perform straight line fitting on the point cloud data; and perform vehicle target recognition based on the straight line fitting results.

Optionally, the category is determined by setting the geometric distance between two adjacent points.

Optionally, the clustering process includes dividing the scattered two-dimensional point cloud data into multiple point classes, wherein the geometric coordinates of points within the point class are similar.

Optionally, the operation of merging multiple categories includes: selecting an interest area, and merging the categories in the interest area.

Optionally, the operation of selecting the region of interest includes: setting the measurement angle and distance according to the application scenario, and selecting the region of interest.

Optionally, the operation of merging multiple categories includes: merging the categories according to vehicle object shape and contour features and the reason why the vehicle is divided into multiple categories.

Optionally, the straight line fitting operation includes: removing breakpoints from the clustered point cloud data to remove outliers.

Optionally, the straight line fitting operation includes: obtaining a straight line or a broken line through straight line fitting.

Optionally, the operation of identifying the vehicle target includes: complementing the straight line or folded line into a complete rectangle; comparing the size of the rectangle with the vehicle target, and screening based on the speed information of the points to complete the recognition of the vehicle target.

Optionally, it also includes: using a Kalman filter tracker to continuously track the identified target.

Therefore, the embodiment of the present invention provides a vehicle target recognition system to solve the problem that the visual sensor method to identify vehicle targets is extremely sensitive to changes in the illumination conditions, weather conditions, and traffic conditions of the road environment, and has poor environmental adaptability. The ultrasonic radar sensor method can reliably identify vehicle targets. The working distance is relatively short for technical problems.

It should be noted that for the foregoing method embodiments, for the sake of simple description, they are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on such an understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products are stored in a storage medium (such as ROM/RAM, disk, CD) contains several instructions to enable a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

The serial numbers of the above embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

In the above-mentioned embodiments of the present invention, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed technical content can be realized in other ways. Wherein, the device embodiments described above are only illustrative, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or can be Integrate into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of units or modules may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present invention. And aforementioned storage medium comprises: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or CD etc. various mediums that can store program codes .

The above is only a preferred embodiment of the present invention, it should be pointed out that, for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications can also be made. It should be regarded as the protection scope of the present invention.

Claims (10)

1. a kind of vehicle target recognition methods characterized by comprising

The point cloud data of ambient enviroment is obtained from mobile lidar；

Clustering processing is carried out to point cloud data, obtains multiple classifications；

The multiple classification is merged；

Straight line fitting is carried out to point cloud data using singular value decomposition algorithm；And

Vehicle target identification is carried out according to straight line fitting result.

2. vehicle target recognition methods according to claim 1, which is characterized in that the classification is by setting adjacent two What the geometric distance of point determined.

3. vehicle target recognition methods according to claim 1, which is characterized in that the clustering processing includes will be at random Two-dimentional point cloud data is divided into multiple classes, wherein the geometric coordinate of described class internal point is close.

4. vehicle target recognition methods according to claim 1, which is characterized in that merged to the multiple classification Operation includes: to choose interest region, and merge to the classification in the interest region.

5. vehicle target recognition methods according to claim 1, which is characterized in that the operation of the straight line fitting includes: Point cloud data after being clustered carries out breakpoint rejecting, removes exceptional value.

6. vehicle target recognition methods according to claim 5, which is characterized in that the operation of the straight line fitting includes: Straight line or broken line are obtained by the straight line fitting.

7. vehicle target recognition methods according to claim 6, which is characterized in that carry out the operation packet of vehicle target identification It includes: the straight line or broken line is mended as complete rectangle；The rectangle is carried out size with vehicle target to compare, and binding site Velocity information is screened, and the identification to vehicle target is completed.

8. a kind of storage medium, which is characterized in that the storage medium includes the program of storage, wherein run in described program When control the storage medium where equipment perform claim require any one of 1 to 7 described in vehicle target recognition methods.

9. a kind of processor, which is characterized in that the processor is for running program, wherein right of execution when described program is run Benefit require any one of 1 to 7 described in vehicle target recognition methods.

10. a kind of vehicle target identifying system characterized by comprising

Processor；And

Memory is connected to the processor, for providing the instruction for handling following processing step for the processor:

The point cloud data of ambient enviroment is obtained from mobile lidar；

Clustering processing is carried out to point cloud data, obtains multiple classifications；

The multiple classification is merged；

Straight line fitting is carried out to point cloud data using singular value decomposition algorithm；And

Vehicle target identification is carried out according to straight line fitting result.