JP7548859B2 - Object Tracking Device - Google Patents

Description

This disclosure relates to an object tracking device.

Conventionally, the most common method for recognizing objects around a vehicle using an on-board radar device is to perform frequency analysis on the observed signal obtained from the radar device and extract the maximum power point of the frequency-analyzed signal (i.e., peak search).

However, depending on the environment around the vehicle, it may not be possible to extract a signal corresponding to a desired object as a peak.
As an effective method in such cases, a technology has been proposed in which a predicted value for the object at the current time is calculated using the results of past object tracking, and a signal indicating the object is extracted based on the predicted value (see, for example, Patent Document 1).

JP 2003-270341 A

However, as a result of detailed investigation by the inventors, the following problems were found in the conventional techniques.
Specifically, when the objects around the vehicle (i.e., the radar target) or the behavior of the vehicle change significantly, the error of the predicted value may become large, and in such a case, the error of the extracted signal (i.e., the history signal) also becomes large.

Therefore, even if normal signals at the current time can be correctly detected, if the predicted value is associated with a historical signal containing an error, the target or the vehicle's behavior cannot be appropriately responded to, and the target may be lost (i.e., lost). In other words, the above-mentioned technology may not be able to track objects accurately.

One aspect of the present disclosure is to provide technology that enables accurate tracking of objects.

One aspect of the present disclosure relates to an object tracking device (3) configured to track targets around a moving object based on observation signals from a sensor (7).
This object tracking device includes an acquisition unit (21), a signal analysis unit (23), a first signal extraction unit (25), a prediction unit (27), a second signal extraction unit (29), a cost setting unit (31), an association unit (33), an estimation unit (35), a reliability determination unit (37), and a cost modification unit (39).

The acquisition unit is configured to acquire the observation signal from the sensor.
The signal analysis unit is configured to generate an intensity distribution of the observed signal or a signal derived from the observed signal.

The first signal extractor is configured to extract one or more first detection signals corresponding to the target based on the intensity distribution.
The prediction unit is configured to calculate a prediction value of the state of the target at a current time based on a past state (e.g., distance, speed, direction) of the target.

The second signal extractor is configured to extract, based on the intensity distribution, one or more second detection signals corresponding to the predicted value of the target.
The cost setting unit is configured to set an association cost between the predicted value and the first detection signal, and an association cost between the predicted value and the second detection signal.

The association unit is configured to find an associated signal that satisfies a condition for association with the predicted value (e.g., a condition that the association cost is minimal) based on the association cost, and associate the predicted value with the associated signal.

The estimation unit is configured to estimate the current state of the target from the predicted value and the associated signal (i.e., a signal associated with the predicted value, e.g., the first detection signal or the second detection signal, or a signal obtained from the first detection signal and/or the second detection signal).

The reliability determining unit is configured to determine the reliability of the second detection signal.
The cost modification unit is configured to modify the association cost between the predicted value and the second detection signal based on a result of the determination by the reliability determination unit.

With this configuration, the present disclosure enables tracking of targets (i.e. objects) with high accuracy.
In other words, in the present disclosure, the association cost between the predicted value and the second detection signal is changed based on the result of determining the reliability of the second detection signal, so that the association in the association unit can be appropriately performed based on the changed association cost. For example, the predicted value of the combination with the lowest association cost can be associated with the associated signal.

Therefore, the current state of the target can be estimated with high accuracy from the predicted value and the related signal (i.e., the signal associated with the predicted value), so that the object can be tracked with high accuracy.
Here, the association cost is an index indicating the degree of association between a combined predicted value and each of the detection signals when the predicted value and each of the detection signals are combined, and more specifically, is an index indicating the difficulty in associating a predicted value with an observed value corresponding to each of the detection signals (i.e., a value indicating the state of the target obtained from the observed signal).

The lower the association cost, the higher the correlation between the predicted value and the observed value (and therefore each detection signal). For example, the lower the association cost, the higher the likelihood that the observed value represents the actual state of the target.

The association refers to a process in which, when estimating the current state of a target based on a predicted value and an associated signal (e.g., each detection signal or a value obtained from each detection signal), the associated signal used for the estimation is found based on a specified condition (e.g., association cost) and the predicted value is linked to the associated signal found under the specified condition.

The related signal is a signal associated with the predicted value, i.e., a signal used together with the predicted value when estimating the current state of the target (i.e., a signal obtained under the specified conditions). Examples of the related signal include the first detection signal and the second detection signal, or a signal obtained from the first detection signal and/or the second detection signal (e.g., a signal obtained by weighting).

Note that the symbols in parentheses in this section and in the claims indicate the correspondence with the specific means described in the embodiments described below as one aspect, and do not limit the technical scope of this disclosure.

1 is a block diagram showing a configuration of an object tracking system including an object tracking device according to a first embodiment. FIG. 2 is a diagram showing an example of a detection area when a radar device is mounted in front of a vehicle. 11 is a diagram showing an example of a detection area when a radar device is mounted in a location other than the front of the host vehicle; FIG. FIG. 2 is a block diagram showing the functional configuration of the object tracking device. FIG. 2 is an explanatory diagram showing the state of reflected radar waves when there is a roadside wall; FIG. 11 is an explanatory diagram showing a method of detecting a normal signal and a history signal. FIG. 11 is an explanatory diagram showing a state of lane change and a state in which a prediction error occurs; FIG. 11 is an explanatory diagram regarding associated costs. 1 is a main flowchart showing a tracking process. 11 is a schematic diagram for explaining a process for calculating a current estimated value from a previous estimated value; FIG. 13 is a flowchart showing a cost setting process. 13 is a flowchart showing a cost addition determination process. 10 is a flowchart showing a process according to a second embodiment.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[1. First embodiment]
[1-1. Overall configuration]
First, the overall configuration of an object tracking system including an object tracking device in the first embodiment will be described.

As shown in FIG. 1, in the object tracking system 1, the object tracking device 3 is a device mounted on a moving body for tracking objects around the moving body. The moving body is, for example, a vehicle, an aircraft, a ship, etc.

In addition to the object tracking device 3, the moving body is equipped with a radar device 7 and various detection devices other than the radar device 7 (i.e., other detection devices) 9 as sensors 5. Examples of the detection devices 9 include well-known vehicle speed sensors, longitudinal acceleration sensors, lateral acceleration sensors, yaw rate sensors, navigation devices, etc.

In this first embodiment, the object tracking device 3 and the sensor 5 are mounted on the host vehicle JV. Note that vehicles other than the host vehicle JV, such as a preceding vehicle, are referred to as other vehicles, but vehicles in general may be simply referred to as vehicles.

As shown in FIG. 2, the radar device 7 may be mounted in the front center of the host vehicle JV (e.g., in the center of the front bumper), and the area in the front center of the host vehicle JV may be the detection area Rd. Also, as shown in FIG. 3, the radar device 7 may be mounted in five locations, namely, the front center, left front side, right front side, left rear side, and right rear side, of the host vehicle JV, and the areas in the front center, left front, right front, left rear, and right rear of the host vehicle JV may be the detection area Rd. The number and mounting positions of the radar devices 7 mounted on the host vehicle JV may be selected as appropriate.

The radar device 7 is a high-resolution millimeter wave radar and includes a transmitting array antenna composed of multiple antenna elements and a receiving array antenna composed of multiple antenna elements. The radar device 7 irradiates a transmission wave to the detection area Rd in each processing cycle that arrives at a predetermined period Tcy.

The radar device 7 then receives the reflected waves (i.e., received waves) that are generated when the transmitted waves are reflected at the reflecting point of the target. Examples of targets include vehicles, road surfaces, roadside objects, and other objects. Furthermore, the radar device 7 generates a beat signal by mixing the transmitted waves and the reflected waves, and outputs the signal generated by sampling the beat signal to the object tracking device 3.

The signal output from the radar device 7 is called the observation signal. Note that, in this example, a signal generated by sampling a beat signal is output as the observation signal, but this is not limited to this. Note that the radar device 7 may use any modulation method, such as the FMCW method or the multi-frequency CW method.

Returning to FIG. 1, the object tracking device 3 is mainly composed of a well-known microcomputer (i.e., a microcomputer) having a CPU 11, a ROM 13, a RAM 15, a flash memory 17, etc. The CPU 11 executes a program stored in the ROM 13. By executing the program, a method corresponding to the program is performed. Note that memories 19 such as the ROM 13, the RAM 15, and the flash memory 17 are non-transitive tangible recording media. The object tracking device 3 may also include a coprocessor that executes fast Fourier transform processing (i.e., FFT processing) and the like.

The number of microcomputers constituting the object tracking device 3 may be one or more. Furthermore, the method of realizing the various functions possessed by the object tracking device 3 is not limited to software, and some or all of the elements may be realized using one or more pieces of hardware. For example, when the above functions are realized by electronic circuits that are hardware, the electronic circuits may be realized by digital circuits including multiple logic circuits, or analog circuits, or a combination of these.

The object tracking device 3 performs a tracking process based on the observation signal generated by the radar device 7, and estimates a state quantity indicating the state of the target. The state quantity of the target may be, for example, the position of the target relative to the host vehicle JV, or the relative speed of the target relative to the host vehicle JV. More specifically, the position of the target may be the distance from the host vehicle JV to the target, and the orientation of the target relative to the host vehicle JV. Instead of the relative speed of the target, the ground speed of the target calculated from the relative speed and the speed of the host vehicle JV may be used as the state quantity indicating the state of the object. The state quantity of the target may also include the reflection intensity from the reflection point of the target.

[1-2. Configuration of Object Tracking Device]
Next, the configuration of the object tracking device 3 will be described functionally.
As shown in Figure 4, the object tracking device 3 has, as its functional configuration, an acquisition unit 21, a signal analysis unit 23, a first signal extraction unit 25, a prediction unit 27, a second signal extraction unit 29, a cost setting unit 31, an association unit 33, an estimation unit 35, a reliability determination unit 37, and a cost modification unit 39.

The acquisition unit 21 is configured to acquire the output signal (i.e., the normal signal) from the radar device 7. In other words, it is configured to acquire the observation signal for each receiving array antenna.

The signal analysis unit 23 is configured to perform FFT processing on the observed signal to generate an intensity distribution (i.e., a power spectrum). As is well known, the peak in this power spectrum corresponds to the distance to the target, i.e., the frequency of the power spectrum corresponds to the distance, so hereinafter this power spectrum will be referred to as a distance power spectrum.

The power spectrum can be obtained by performing FFT processing on the signal components extracted from each receiving array antenna at the same frequency. As is well known, the frequency of this power spectrum corresponds to the direction, so below this power spectrum will be referred to as the direction power spectrum.

The first signal extraction unit 25 is configured to extract one or more first detection signals corresponding to a target based on the intensity distribution (e.g., a distance power spectrum or a distance-velocity power spectrum described later) obtained from a normal signal. Note that this first detection signal corresponds to an observation value (i.e., a first observation value) indicating the position, etc., of the target (see, for example, FIG. 8).

The prediction unit 27 is configured to calculate a predicted value of the state of the target at the current time based on an estimated value of the state of the target in the past.
For example, from an estimated value of the target's state quantity in the previous cycle (e.g., target information such as distance from the target, relative speed from the target, and direction of the target from the host vehicle JV), a predicted value of the target's state quantity at the current time (e.g., the current cycle) (e.g., predicted values of the target's distance from the target, relative speed from the target, and direction of the target from the host vehicle JV at the current time) is calculated.

In the following, the distance to the target, the relative speed to the target, and the direction of the target from the host vehicle JV may be simply referred to as distance, speed, and direction, respectively.
The second signal extraction unit 29 is configured to extract one or more second detection signals corresponding to a predicted value of a target based on the intensity distribution (for example, a distance power spectrum or a distance-velocity power spectrum).

That is, the second detection signal is extracted by taking into consideration the intensity distribution obtained from the normal signal and the predicted value obtained from the past estimated value. Note that this second detection signal corresponds to the observation value (i.e., the second observation value) indicating the position of the target, etc. (see, for example, FIG. 8).

For example, when distance and speed information is available as predicted values, all cells within a predetermined distance and speed range for the predicted values in the distance-speed power spectrum may be extracted, and the maximum power peak in those cells may be used as the second detection signal.

Alternatively, direction estimation may be performed for all cells extracted as described above, and the peak in the cell closest to the predicted value may be determined as the second detection signal.
The cost setting unit 31 is configured to set an association cost between the predicted value and the first detection signal, and an association cost between the predicted value and the second detection signal.

Here, the association cost is an index showing the difficulty of associating the predicted value with each of the first and second detection signals, and the lower the association cost, the higher the association between the predicted value and each of the detection signals. Therefore, the lower the association cost, the easier it is to associate the predicted value with each of the detection signals.

For example, if the distance between the predicted value and the observed value corresponding to each detection signal (e.g., Euclidean distance or Mahalanobis distance) is used as the association cost, the smaller the distance between the predicted value and the observed value, the smaller the association cost.

When distance is used to calculate the association cost below, Euclidean distance, Mahalanobis distance, or the like can be adopted.
The associating unit 33 is configured to find an associated signal that satisfies a condition for associating with the predicted value (for example, a condition that the association cost is the smallest) based on the association cost, and associate the predicted value with the associated signal.

For example, the cost of associating the predicted value with the first detection signal is compared with the cost of associating the predicted value with the second detection signal, and the detection signal with the lowest association cost is adopted as the associated signal. Therefore, the associated signal is associated with the predicted value. Note that in this case, the detection signal associated with the predicted value is the associated signal.

As described below, when the cost modification unit 39 modifies the association cost between the predicted value and the second detection signal, the association is performed based on the modified association cost.

This association is a process for selecting a current detection signal (i.e., a related signal) that allows for the calculation of a more accurate current estimation value when, for example, calculating a current estimation value from a current predicted value and the current detection signals of the first detection signal and the second detection signal.

For example, if the distance between the predicted value and the observed value corresponding to each detection signal is used as the association cost, the detection signal that is the closest to the predicted value can be associated with the predicted value.

As described below, when there are multiple first detection signals and/or multiple second detection signals, a related signal may be obtained from the predicted value and the first detection signal and/or the detection signal based on the association cost, and the related signal may be associated with the predicted value.

The estimation unit 35 is configured to estimate the current state of the target (e.g., state quantities such as distance, speed, and direction) from the predicted value and a signal associated with the predicted value (e.g., a related signal such as the first detection signal or the second detection signal).

The reliability determination unit 37 is configured to determine the reliability of the second detection signal. For example, it is configured to determine the reliability of the second detection signal based on an output signal from the sensor 5 indicating the state of the host vehicle JV and the preceding vehicle, etc.

The cost modification unit 39 is configured to modify the cost of associating the predicted value with the second detection signal based on the determination result of the reliability determination unit 37 .
[1-3. Processing Principle]
Next, the principle of the method for tracking a target in the object tracking device 3 will be described.

As shown in Figure 5, for example, when the host vehicle JV and a preceding vehicle, which is a target, are traveling along the side wall of the road, if the host vehicle JV irradiates radar waves onto the preceding vehicle, the reflected waves of the radar waves may follow different paths.

Specifically, the host vehicle JV may directly receive radar waves reflected by a preceding vehicle (see direct reflection in Figure 5), or may receive radar waves that are reflected by the roadside wall after being reflected by the preceding vehicle's reflected waves (see multipath in Figure 5).

In such a situation, if the received radar wave is subjected to FFT processing to obtain the distance power spectrum, a signal that is a combination of the direct reflection signal and the multipath signal (i.e., the observed signal after FFT processing) is obtained, as shown in the upper diagram of Figure 6. Therefore, if the azimuth power spectrum is further obtained based on the peak of this FFT processed signal, there is a risk of errors occurring in the azimuth estimation.

In FIG. 6, the normal signal refers to the signal at the current time (e.g., the signal obtained in the current cycle of processing), and the history signal refers to the signal obtained in the past (e.g., the signal obtained in the previous cycle of processing).

As a countermeasure, in a situation where the above-mentioned multipath occurs, as shown in the lower diagram of Figure 6, information on the likelihood that a target exists (e.g., the predicted distance between the host vehicle JV and the target) is obtained based on the predicted value of the state quantity of the target obtained from past signals (i.e., historical signals), and a signal corresponding to this predicted distance is extracted. Note that here, the predicted distance is considered to correspond to the peak obtained from the direct reflection signal. Then, if the heading power spectrum is obtained based on this extracted signal, it is highly likely that errors in heading estimation can be suppressed.

In other words, it is believed that by using the history signal, it is possible to accurately detect the position of a target including its orientation, that is, to accurately track an object.
However, as shown in the upper diagram of Figure 7, when the driving state suddenly changes, for example when the host vehicle JV changes lanes, another problem may occur.

Specifically, as shown in the lower diagram of Figure 7, when the driving conditions, etc. suddenly change, if the predicted value of the target position, etc. is calculated based on the history signal, there is a risk of a prediction error occurring compared to when only normal signals are used. This is because a sudden change in the driving conditions, etc. is thought to affect the state of the reflected radar waves, etc.

Therefore, in this first embodiment, taking into consideration the fact that a highly accurate prediction value may not be obtained, such as in the case of a sudden change in the driving state, the reliability of the second detection signal (i.e., a signal that takes into account the history signal in addition to the normal signal) is determined, and the association cost between the prediction value and the second detection signal is changed based on the determination result.

In more detail, when the reliability determination unit 37 determines that a predetermined condition indicating that the reliability of the second detection signal is low is satisfied, the cost of associating the predicted value with the second detection signal is increased so that the association becomes more difficult than when the predetermined condition is not satisfied.

For example, as shown in the upper diagram of Figure 8, when the first detection signal and the second detection signal are obtained within a predetermined prediction range (e.g., a prediction gate described below) for a predicted value (e.g., a current predicted value), the association cost (cost) between the predicted value and the first detection signal is 2, and the association cost (cost) between the predicted value and the second detection signal is 1 when the reliability of the second detection signal is not taken into consideration.

In FIG. 8, the x mark of the first detection signal corresponds to the observation value of the state quantity of the target obtained from the normal signal (e.g., the first observation value), and the x mark of the second detection signal corresponds to the observation value of the state quantity of the target obtained by taking the history signal into account in the normal signal (e.g., the second observation value).

In this first embodiment, in such a case, as shown in the lower diagram of Figure 8, by taking into account the reliability of the second detection signal, it is possible to change the association cost between the predicted value and the second detection signal to a larger value (e.g., to 11) than when the reliability of the second detection signal is not taken into account.

In such a case, that is, when the association cost between the predicted value and the second detection signal becomes greater than the association cost between the predicted value and the first detection signal, the predicted value and the first detection signal, which are a combination with a smaller association cost, are associated with each other. In other words, the first detection signal is adopted as an associated signal to associate the predicted value with the first detection signal.

In addition, by taking into account the reliability of the second detection signal, if the association cost between the predicted value and the second detection signal becomes smaller than 2, the predicted value with the smaller association cost is associated with the second detection signal.

Therefore, by estimating the current state of the target (i.e., obtaining the current estimated value, which will be described later) based on the associated predicted value and each detection signal (i.e., the associated signal), it is possible to track the object with high accuracy.

In this way, when the reliability of the second detection signal is high, the object can be tracked with high accuracy based on the second detection signal, whereas when the reliability of the second detection signal is low, the object can be tracked with high accuracy based on the first detection signal.

As a result, even if there is a sudden change in the driving conditions, the object can always be tracked with high accuracy.
[1-4. Processing Overview]
The tracking process executed by the CPU 11 of the object tracking device 3 will be described with reference to the flowchart shown in Fig. 9. This process is repeatedly started for each processing cycle.

First, in S100, the CPU 11 acquires a signal (for example, an observation signal) corresponding to each target output from the radar device 7.
Next, in S110, the CPU 11 executes a distance power spectrum calculation process, that is, performs FFT processing on the observed signal to calculate a distance power spectrum, as described above.

Note that, as is well known, the speed power spectrum may be obtained by further performing FFT processing on the distance power spectrum. This makes it possible to obtain a two-dimensional power spectrum of distance and speed (i.e., a distance-speed power spectrum).

Furthermore, as described above, the azimuth power spectrum may be calculated to obtain information on distance, speed, and azimuth.
Next, in S120, the CPU 11 executes a prediction process. In the first embodiment, the CPU 11 calculates a current predicted value, which is a predicted value of the state quantity of the target in the current processing cycle, from a previous estimated value, which is an estimated value of the state quantity of the target calculated in the previous processing cycle.

As an example, FIG. 10 shows the target position indicated by the previous estimated value, the target position indicated by the current predicted value, and the target position indicated by the current estimated value. Note that the cross marks with diagonal lines inside in FIG. 10 indicate state quantities such as position obtained from the normal signal in the current cycle (i.e., the first observation value corresponding to the first detection signal), and the black cross marks indicate state quantities such as position obtained by taking the normal signal into account with the historical signal in the current cycle (i.e., the second observation value corresponding to the second detection signal).

The CPU 11 also sets a prediction gate based on the current predicted value. It is believed that the position indicated by each observation value corresponding to the same target as the target corresponding to the current predicted value is near the position indicated by the current predicted value. Therefore, an area centered on the position indicated by the current predicted value is set as a prediction gate, which is an area in which it is estimated that the position indicated by each observation value corresponding to the same target as the current predicted value is present.

As an example, FIG. 10 shows a circular prediction gate that is set for the position indicated by the current prediction value. For example, a prediction gate showing a size according to distance (e.g., Euclidean distance) is shown. Note that the size of the prediction gate may be determined according to, for example, the relative speed indicated by the state quantity.

Next, in S130, the CPU 11 executes a first signal extraction process. For example, a first detection signal whose peak exceeds a threshold and is a local maximum point is extracted from the distance power spectrum. Note that a first detection signal whose peak exceeds a threshold and is a local maximum point may also be extracted from the distance speed power spectrum.

Next, in S140, the CPU 11 executes a second signal extraction process. For example, a distance power spectrum is used to extract a second detection signal corresponding to a predicted value (e.g., distance) of the target. Alternatively, a distance-speed power spectrum may be used to extract a second detection signal corresponding to a predicted value (e.g., distance, speed) of the target.

In more detail, for example, a peak closest to the distance of the predicted value may be extracted based on the distance power spectrum, and the value indicated by that peak (i.e., the value corresponding to the distance) may be extracted as the second detection signal.

As described above, a cell corresponding to a predicted value (e.g., distance, speed) may be selected in the distance-speed power spectrum, and cells in the vicinity of the selected cell may be searched for, and a signal corresponding to the cell with the maximum power may be extracted as the second detection signal.

Next, in S150, the CPU 11 executes a cost setting process. This cost setting process is a process for setting an association cost between the predicted value and the first detection signal, and an association cost between the predicted value and the second detection signal. This cost setting process will be described in detail later.

Next, in S160, the CPU 11 executes an association process. In this first embodiment, this association process is a process of associating the predicted value with the first detection signal or the second detection signal based on the association cost described above (e.g., the distance). In other words, of the first detection signal or the second detection signal, the detection signal with the smaller association cost is adopted as the associated signal, and the associated signal is associated with the predicted value.

For example, if the association cost is set according to the distance, the predicted value is associated with the detection signal that is closer to the predicted value than the first detection signal or the second detection signal, based on the distance between the predicted value and the first detection signal or the second detection signal.

As will be described later, the predicted value may be associated with a signal obtained from the first detection signal and/or the second detection signal.
Next, in S170, the CPU 11 executes the estimation process and temporarily ends this process.

This estimation process involves calculating current estimates of the state quantities of each target in the current processing cycle using a Kalman filter or the like.
As an example, in S160, if there is an observation value (i.e., the first detection signal or the second detection signal) associated with the current predicted value in the current processing cycle, the CPU 11 calculates a current estimated value of the state quantity of the target based on the current predicted value and the observation value associated with the current predicted value. As a result, for example, as shown in Fig. 10, the current estimated value is calculated based on the current predicted value and the observation value associated with the current predicted value (e.g., the second detection signal indicating the second observation value).

[1-5. Cost setting process]
Next, the cost setting process will be described with reference to the flowchart shown in FIG.
First, in S200, the CPU 11 judges whether or not there is a predicted value that has not yet been processed (for example, a current predicted value). Specifically, the CPU 11 judges whether or not there is a current predicted value that has not yet been processed in the subsequent steps S210 to S260 among the current predicted values. If the judgment here is affirmative, the process proceeds to S210, whereas if the judgment here is negative, the process is temporarily terminated.

If the determination here is affirmative, one of the unprocessed predicted values is selected and the process proceeds to S210.
In S210, the CPU 11 determines whether there is an unprocessed observed value. In other words, it determines whether there is an unprocessed observed value among the observed values indicating the position in the prediction gate corresponding to the currently selected current predicted value, i.e., the first observed value and the second observed value in the prediction gate. The unprocessed observed value means an observed value that has not been associated with the currently selected current predicted value and has not had its cost calculated by the subsequent processing from S220 onward.

If the determination here is affirmative, one of the unprocessed observed values is selected and the process proceeds to S220.
In S220, the CPU 11 calculates a cost (i.e., an association cost) between the current predicted value selected in S200 and one observed value selected in S210. Specifically, the CPU 11 calculates a smaller cost as the distance (e.g., Euclidean distance) between the position indicated by the current predicted value and the position indicated by the observed value becomes shorter.

When calculating the association cost between the current predicted value and the observed value, for example, the Mahalanobis distance using distance (i.e., Euclidean distance), speed, and direction as parameters may be used.
In the next step S230, the CPU 11 determines whether the observed value is calculated from the second detection signal. If the determination is affirmative, the process proceeds to step S240. If the determination is negative, the process returns to step S210.

In S240, the CPU 11 performs a cost addition determination process. This cost addition determination process is a process for determining whether or not to add a cost depending on the reliability of the second detection signal. This cost addition determination process will be described in detail later.

In the next step S250, the CPU 11 determines whether or not a cost is to be added. In other words, this is a process for determining whether or not to add a cost depending on the result of the cost addition determination process in S240, and therefore depending on the reliability of the second detection signal. If the determination here is affirmative, the process proceeds to S260, whereas if the determination here is negative, the process returns to S210.

In S260, the CPU 11 performs the cost addition and returns to S210. In other words, since it was determined in S250 that the current observation value is subject to cost addition, the necessary cost according to the result of the cost addition determination process is added to the cost calculated in S210.

[1-6. Cost addition determination process]
Next, the cost addition determination process will be described with reference to the flowchart shown in FIG.

First, in S300, the CPU 11 determines whether or not the lateral speed of the host vehicle JV is equal to or greater than a threshold value. If the determination is affirmative, the process proceeds to S380, whereas if the determination is negative, the process proceeds to S310.
In addition, lateral speed refers to the left-right speed perpendicular to the fore-and-aft direction of the host vehicle JV when the host vehicle JV is viewed from a direction perpendicular to the horizontal plane, and can be obtained from the output signal from the detection device 9 of the host vehicle JV (e.g., a vector indicating the direction of movement of the host vehicle JV).

In S310, the CPU 11 determines whether the lateral speed of the target is equal to or greater than a threshold. If the answer is yes, the process proceeds to S380. If the answer is no, the process proceeds to S320. The lateral speed of the object can be calculated from past data on the state quantities of the target.

In S320, the CPU 11 determines whether the acceleration (e.g., longitudinal acceleration) of the host vehicle JV is equal to or greater than a threshold value. If the answer is affirmative, the process proceeds to S380; if the answer is negative, the process proceeds to S330. The acceleration of the host vehicle JV can be determined by a longitudinal acceleration sensor.

In S330, the CPU 11 determines whether the acceleration of the target (e.g., longitudinal acceleration) is equal to or greater than a threshold value. If the determination is affirmative, the process proceeds to S380, whereas if the determination is negative, the process proceeds to S340. The acceleration of the object can be calculated from past data on the state quantities of the target.

In S340, the CPU 11 determines whether the radius of curvature of the road shape is equal to or greater than a threshold value. If the answer is yes, the process proceeds to S380; if the answer is no, the process proceeds to S350. The radius of curvature of the road shape can be obtained from data on the past state quantities of the preceding vehicle (e.g., data on the trajectory of a target object in the second embodiment described below), data on the past state quantities of the host vehicle JV (e.g., data on the trajectory of the host vehicle JV), data on the position of roadside objects such as guardrails, etc.

In S350, the CPU 11 determines whether the power of the signal (e.g., the second detection signal) is equal to or greater than a threshold value. If the determination is affirmative, the process proceeds to S380, whereas if the determination is negative, the process proceeds to S360. Note that, in addition to the second detection signal, various types of signals obtained from reflected waves of radar waves may be used as this signal.

In S360, the CPU 11 determines whether the average value of the prediction errors up to the previous cycle is equal to or greater than a threshold value. If the answer is yes, the process proceeds to S380; if the answer is no, the process proceeds to S370.

The average value of the prediction error up to the previous cycle may be the average value of the prediction error over multiple cycles. In addition, examples of the prediction error include the difference between the current predicted value and the observed value (e.g., the observed value corresponding to the associated signal) and the difference between the current predicted value and the current estimated value.

In S370, the CPU 11 determines that the observation value currently being judged is not a cost addition target, and for example sets a predetermined flag, and temporarily ends this process.
On the other hand, in S380, the CPU 11 determines that the observation value currently being judged is a target for cost addition, sets a predetermined flag, for example, and temporarily ends this process.

In addition, in steps S300 to S360, if a positive judgment is made, it is determined that a cost should be added, and the amount of cost to be set in such a case is predetermined.

For example, the same cost (e.g., 10) may be added when a positive judgment is made at each step. Alternatively, different costs may be added when a positive judgment is made at each step, distinguishing between conditions in which errors are likely to occur.

[1-7. Effects]
In the above-described first embodiment, the following advantageous effects can be obtained.
(1a) In the first embodiment, the state quantity of the current target (i.e., an object such as a preceding vehicle) can be accurately estimated from the predicted value and an associated signal associated with the predicted value, so that the object can be tracked with high accuracy.

That is, in the first embodiment, the cost of associating the predicted value with the second detection signal can be changed based on the result of determining the reliability of the second detection signal.
For example, when the reliability determination unit 37 determines that a predetermined condition indicating that the reliability of the second detection signal is low is satisfied, the cost of associating the predicted value with the second detection signal can be set to a high value so that association becomes more difficult than when the predetermined condition is not satisfied.

Therefore, the association in the association unit 33 can be appropriately performed based on this changed association cost. Therefore, the current state quantity of the target can be accurately estimated from the predicted value and the detection signal associated with the predicted value (i.e., the associated signal), so that the object can be tracked with high accuracy.

(1b) In this first embodiment, the first signal extraction unit 25 can extract a first detection signal whose intensity exceeds a threshold and is a maximum point from the intensity distribution of the observation signal (e.g., distance power spectrum). In addition, the second signal extraction unit 29 can extract a second detection signal corresponding to a predicted value of a target from the intensity distribution of the observation signal.

(1c) In this first embodiment, when a first condition is satisfied in which the absolute value of the lateral speed of the host vehicle JV or the target is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost can be set to be large so that association is more difficult than when the first condition is not satisfied.

(1d) In this first embodiment, when the second condition is satisfied, that is, the absolute value of the acceleration of the host vehicle JV or the target is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost can be set to be large so that association is more difficult than when the second condition is not satisfied.

(1e) In this first embodiment, when the third condition is satisfied, that is, the radius of curvature of the road shape is smaller than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost can be set to be large so that association is more difficult than when the third condition is not satisfied.

(1f) In the first embodiment, when a fourth condition is satisfied in which the power of the signal obtained from the radar device 7 (e.g., the power of the second detection signal) is less than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost can be set to be high so that association is more difficult than when the fourth condition is not satisfied.

(1g) In the first embodiment, when the fifth condition is satisfied, that is, the past prediction error is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost can be set to be large so that association is more difficult than when the fifth condition is not satisfied.

[1-8. Correspondence of Wording]
In the relationship between this first embodiment and the present disclosure, the radar device 7 corresponds to the sensor, the object tracking device 3 corresponds to the object tracking device, the acquisition unit 21 corresponds to the acquisition unit, the signal analysis unit 23 corresponds to the signal analysis unit, the first signal extraction unit 25 corresponds to the first signal extraction unit, the prediction unit 27 corresponds to the prediction unit, the second signal extraction unit 29 corresponds to the second signal extraction unit, the cost setting unit 31 corresponds to the cost setting unit, the association unit 33 corresponds to the association unit, the estimation unit 35 corresponds to the estimation unit, the reliability determination unit 37 corresponds to the reliability determination unit, and the cost change unit 39 corresponds to the cost change unit.

[2. Second embodiment]
Since the second embodiment has a basic configuration similar to that of the first embodiment, the following description will mainly focus on the differences from the first embodiment. Note that the same reference numerals as those in the first embodiment indicate the same configurations, and the preceding description will be referred to.

Since the second embodiment differs from the first embodiment in the content of the tracking process, the tracking process will be described below.
As shown in FIG. 13, the second embodiment performs the same processes as S100 to S170 in the first embodiment.

Specifically, in S400, the same process of acquiring observed signals as in S100 is performed, in S410, the same process of calculating distance power spectrum as in S110 is performed, in S420, the same prediction process as in S120 is performed, in S430, the same first signal extraction process as in S130 is performed, in S440, the same second signal extraction process as in S140 is performed, in S450, the same cost setting process as in S150 is performed, in S460, the same association process as in S160 is performed, and in S470, the same estimation process as in S170 is performed.

Then, in the next S480, a road shape estimation process is performed in which an estimate of the road shape is calculated using the estimated values of the state quantities of each target obtained in the above estimation process.
That is, the trajectory of the target can be known from estimated values of the state quantities of the target (for example, distance, speed, direction) and the like, and the road shape can be estimated from this trajectory of the target.

The road shape estimated in this manner can be used, for example, to determine the radius of curvature of the road shape in S340 of FIG.
The second embodiment has the same effects as the first embodiment. Furthermore, the second embodiment can estimate the road shape from the estimated values of the state quantities of the physical conditions.

[3. Third embodiment]
Since the third embodiment has a basic configuration similar to that of the first embodiment, the following description will mainly focus on the differences from the first embodiment. Note that the same reference numerals as those in the first embodiment indicate the same configurations, and the preceding description will be referred to.

In the third embodiment, the processing contents of the first detection signal and the second detection signal are different, so the different points will be mainly described.
In the third embodiment, a related signal is obtained from a plurality of detection signals, ie, the first detection signal and/or the second detection signal, and the related signal is associated with a predicted value.

For example, when one or more first detection signals and one or more second detection signals are extracted within a prediction gate, each detection signal is weighted to calculate an overall detection signal (i.e., a related signal) that is associated with the predicted value.

For example, for multiple detection signals, the association cost with the predicted value is calculated for each, and the association cost is used for weighting, for example, a weighted average is calculated to calculate an overall detection signal. Here, weighting can be performed so that the weight decreases as the association cost increases. Note that the inverse of the association cost may be used as the weighting method.

Then, this overall detection signal is adopted as the related signal. In other words, this related signal is associated with the predicted value. Then, as in the first embodiment, the current state of the target is estimated from the predicted value and the related signal.

In addition, among the detection signals in the prediction gate, multiple detection signals that satisfy a predetermined condition (for example, detection signals whose distance from the predicted value is smaller than a predetermined value) may be selected and weighted to obtain related signals.

The third embodiment has the same effects as the first embodiment. Furthermore, the third embodiment can suitably associate the first detection signal and the second detection signal with a predicted value even when there are multiple first detection signals and multiple second detection signals.

4. Other embodiments
Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments and can be implemented in various modified forms.

(4a) In the first embodiment, for example, if multiple first detection signals are extracted within the prediction gate, the multiple first detection signals may be weighted, for example, using the association cost as in the third embodiment, to obtain an overall first detection signal.

For example, a comprehensive first detection signal may be obtained by taking a weighted average of a plurality of first detection signals, and this comprehensive first detection signal may be used as the related signal.
The same applies to the case where there are a plurality of second detection signals.

(4b) As the association cost, a value based on the distance (e.g., Euclidean distance) between the predicted value and the first detection signal or the second detection signal can be used. In addition, a value based on the Mahalanobis distance using at least two or more of distance, speed, and direction as parameters can be used.

(4c) In the above embodiment, the association cost is described as an index showing the difficulty of associating a predicted value with each detection signal. In other words, the higher the association cost, the lower the association. However, the opposite index can also be used.

For example, instead of the association cost, an index indicating the ease of associating a predicted value with each detection signal can be used. In this case, the larger the index, the stronger the association, so the detection signal with the larger index can be used as the associated signal.

(4d) In the above embodiment, an example was shown in which the radar device adopted the FMCW method, but the radar method of the radar device is not limited to FMCW and may be configured to adopt, for example, two-frequency CW, FCM, or pulse. FCM is an abbreviation for Fast-Chirp Modulation.

(4e) The object tracking device and method described herein may be implemented by a special purpose computer provided by configuring a processor and memory programmed to perform one or more functions embodied in a computer program.

Alternatively, the object tracking device and method described in this disclosure may be implemented by a special-purpose computer provided by configuring a processor with one or more dedicated hardware logic circuits.

Alternatively, the object tracking device and method described in this disclosure may be implemented by one or more special-purpose computers configured by combining a processor and memory programmed to perform one or more functions with a processor configured by one or more hardware logic circuits.

The computer program may also be stored in a computer-readable non-transient tangible recording medium (i.e., a non-transient tangible recording medium) as instructions executed by a computer. The method for realizing the functions of each unit included in the control unit does not necessarily have to include software, and all of the functions may be realized using one or more pieces of hardware.

(4f) Multiple functions possessed by one component in the above embodiments may be realized by multiple components, or one function possessed by one component may be realized by multiple components. Also, multiple functions possessed by multiple components may be realized by one component, or one function realized by multiple components may be realized by one component. Also, part of the configuration of the above embodiments may be omitted. Also, at least part of the configuration of the above embodiments may be added to or substituted for the configuration of another of the above embodiments.

(4g) In addition to the object tracking device described above, the present disclosure can also be realized in various forms, such as a system that includes the object tracking device as a component, a program for causing the computer of the object tracking device to function, a non-transient tangible recording medium such as a semiconductor memory on which the program is recorded, and a control method.

3: Object tracking device, 7: Radar device, 21: Acquisition unit, 23: Signal analysis unit, 25: First signal extraction unit, 27: Prediction unit, 29: Second signal extraction unit, 31: Cost setting unit, 33: Association unit, 35: Estimation unit, 37: Reliability determination unit, 37: Cost change unit

Claims (11)

An object tracking device (3) configured to track a target around a moving object based on an observation signal from a sensor (7),
an acquisition unit (21) configured to acquire the observation signal from the sensor;
a signal analysis unit (23) configured to generate an intensity distribution of the observed signal or a signal derived from the observed signal;
a first signal extracting unit (25) configured to extract one or more first detection signals corresponding to the target based on the intensity distribution;
a prediction unit (27) configured to calculate a prediction value of a state of the target at a current time based on a past state of the target;
a second signal extractor (29) configured to extract one or more second detection signals corresponding to the predicted value of the target based on the intensity distribution;
a cost setting unit (31) configured to set an association cost between the predicted value and the first detection signal and an association cost between the predicted value and the second detection signal;
an association unit (33) configured to obtain an associated signal that satisfies a condition for associating the predicted value with the associated signal based on the association cost between the predicted value and the first detection signal and the association cost between the predicted value and the second detection signal, and associate the predicted value with the associated signal;
an estimation unit (35) configured to estimate a current state of the target from the predicted value and the related signal;
a reliability determination unit (37) configured to determine the reliability of the second detection signal;
a cost modification unit (39) configured to modify the association cost between the predicted value and the second detection signal based on a determination result of the reliability determination unit;
An object tracking device comprising: The object tracking device according to claim 1 ,
When the reliability determination unit determines that a predetermined condition indicating that the reliability of the second detection signal is low is satisfied, the cost of associating the prediction value with the second detection signal is set to be larger than that when the predetermined condition is not satisfied.
Object tracking device. The object tracking device according to claim 1 or 2,
The association unit is configured to associate the predicted value with the first detection signal or the second detection signal based on the association cost.
Object tracking device. The object tracking device according to claim 1 or 2,
The associating unit is configured to obtain the associated signal from the first detection signal and/or the second detection signal based on the association cost, and associate the predicted value with the associated signal.
Object tracking device. The object tracking device according to claim 4,
In the case where there are a plurality of the first detection signals and/or the second detection signals,
The method is configured to weight the plurality of detection signals based on the association cost, and to obtain the associated signal based on the weighted plurality of detection signals.
Object tracking device. The object tracking device according to any one of claims 1 to 5,
the first signal extraction unit is configured to extract, from the intensity distribution, the first detection signal, the intensity of which exceeds a threshold and is a maximum point in the intensity distribution;
The second signal extraction unit is configured to extract the second detection signal corresponding to the predicted value of the target from the intensity distribution.
Object tracking device. The object tracking device according to any one of claims 1 to 6,
When a first condition is satisfied that an absolute value of a lateral speed of the vehicle or the target is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost between the predicted value and the second detection signal is set to be larger than that when the first condition is not satisfied.
Object tracking device. The object tracking device according to any one of claims 1 to 7,
When a second condition is satisfied, that is, an absolute value of the acceleration of the vehicle or the target is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost between the predicted value and the second detection signal is set to be larger than that when the second condition is not satisfied.
Object tracking device. The object tracking device according to any one of claims 1 to 8,
When a third condition is satisfied, that is, a radius of curvature of a road shape is smaller than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost between the predicted value and the second detection signal is set to be larger than that when the third condition is not satisfied.
Object tracking device. The object tracking device according to any one of claims 1 to 9,
When a fourth condition is satisfied, that is, the power of the second detection signal is smaller than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost between the prediction value and the second detection signal is set to be larger than that when the fourth condition is not satisfied.
Object tracking device. The object tracking device according to any one of claims 1 to 10,
When a fifth condition is satisfied, in which a past prediction error is greater than a predetermined value, the reliability of the second detection signal is determined to be low, and the association cost between the prediction value and the second detection signal is set to be larger than when the fifth condition is not satisfied.
Object tracking device.