A target tracking method, computer device, readable storage medium and motor vehicle

Abstract

The application discloses a target tracking method, a computer device, a readable storage medium and a motor vehicle, and belongs to the field of automatic driving technology, and is used for target tracking of a vehicle with an automatic driving function when the vehicle is automatically driven. The target tracking method comprises the following steps: various sensors perceive a tracked target to obtain a measurement of the tracked target; a correlation cost between the measurement of the tracked target and a current state prediction value of the tracked target is calculated; a correlation cost matrix is formed according to the correlation cost, and the measurement of the tracked target and a fused track are correlated according to the correlation cost matrix; the state of the tracked target is updated by filtering; target tracking of a current frame is ended; and target tracking of a next frame is performed. The application can reduce false correlation between different sensor measurements, so that more accurate and stable tracking of a target is achieved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically to a target tracking method, a computer device, a readable storage medium, and a motor vehicle. Background Technology

[0002] Autonomous driving relies on tracking and judging the vehicle ahead. Because single sensors have inherent limitations under specific environmental conditions, they struggle to meet the stringent requirements of intelligent driving systems for robustness and stability. Therefore, multi-target tracking has become a common solution in intelligent driving perception systems. Multi-target tracking refers to using multiple sensors, such as cameras, LiDAR, and millimeter-wave radar, to perceive moving and static targets in autonomous driving scenarios, and then fusing the perception results from each sensor to obtain more accurate and confident tracking results. Therefore, to achieve accurate and stable target tracking, it is necessary to correlate the perception results of the same target from different sensors at different times.

[0003] However, due to environmental interference, sensor performance, and ranging algorithms, perception results such as position, speed, and yaw angle can deviate significantly in real-world scenarios. Furthermore, the different imaging principles of various sensors introduce substantial discrepancies in coordinate system calibration between multiple sensors. The combined effect of these factors leads to significant deviations in measurements of the same target by different sensors at the same time, further impacting the performance of the association algorithm and resulting in frequent false associations in the autonomous driving system. Such false associations are even more frequent in scenarios with densely packed targets. Frequent false associations can cause convergence difficulties or even divergence in the filtering updates of target states. Summary of the Invention

[0004] This invention aims to address one of the technical problems in related technologies to a certain extent. To this end, this invention provides a target tracking method that reduces erroneous correlations between measurements from different sensors, thereby achieving more accurate and stable target tracking.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A target tracking method is provided for a vehicle with autonomous driving capabilities to track targets during autonomous driving. The target tracking method includes the following steps iteratively:

[0007] Various sensors installed on the vehicle perceive the tracked target and acquire measurements of the tracked target;

[0008] Calculate the association cost between the measurement of the tracked target and the predicted current state of the tracked target;

[0009] The correlation cost is used to form a correlation cost matrix, and the measurement and fused trajectory of the tracked target are correlated based on the correlation cost matrix;

[0010] The state of the tracked target is updated by filtering, the target tracking of the current frame ends, and the target tracking of the next frame begins.

[0011] Optionally, the current state prediction of the tracked target can be obtained according to the following formula:

[0012]

[0013] Where k is time k, k+1 is the next time k, and S k+1 S is the predicted value of the tracked target at time k in the next time step. k Let v be the state value of the tracked target at time k. k Let θ be the velocity of the tracked target. k Let θ' be the yaw angle of the tracked target at time k. k Let Δt be the turning speed of the tracked target at time k, and let Δt be the time interval between time k and time k+1.

[0014] Optionally, the association cost between the measurement of the tracked target and the predicted current state of the tracked target is calculated according to the following formula:

[0015] Cost l -1.0×|x l -x k |+0.0×|y l -y k |+0.0×|Vx l -Vx k |+0.0×|Vy i -Vy k |

[0016] Cost v =1.0×|x v -x k |+1.4×|y v -y k |+0.2×|Vx v -Vx k |+0.2×|Vy v -Vy k |

[0017] Cost r =1.0×|x r -x k |+0.7×|y r -y k |+1.0×|Vx r -Vxk |+1.0×|Vy r -Vy k |

[0018] Where Cost is the association cost, x is the longitudinal position of the tracked target in the global coordinate system, y is the lateral position of the tracked target in the global coordinate system, Vx is the longitudinal velocity of the tracked target in the global coordinate system, and Vy is the longitudinal velocity of the tracked target in the global coordinate system; the subscript k indicates that the corresponding parameter at time k is the true value, the subscript l indicates that the corresponding parameter at time k is the measurement of the lidar, the subscript v indicates that the corresponding parameter at time k is the measurement of the vision sensor, and the subscript r indicates that the corresponding parameter at time k is the measurement of the millimeter-wave radar.

[0019] Optionally, before calculating the association cost between the measurement of the tracked target and the predicted current state of the tracked target, the measurement of the tracked target is screened. The screening process for calculating the association cost between the selected measurement of the tracked target and the predicted current state of the tracked target includes the following steps:

[0020] Traverse the merged track of the tracked target;

[0021] Query whether there are any associated measurements in the previous frame of the fused track of the tracked target. If yes, proceed to the next step. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target.

[0022] In the current frame of the fused track of the tracked target, query whether there is a measurement with the same sensor ID as the already associated measurement. If yes, calculate the association cost between the measurement of the tracked target and the current state prediction value of the tracked target. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target.

[0023] Optionally, calculate the correlation probability between the measurement of the tracked target in the current frame and the fused track of the tracked target:

[0024]

[0025] Where, β jt (k) represents the association probability between the j-th measurement and the t-th target, P{δ i (k)|Z k} represents the i-th joint event δ at time k. i The conditional probability of (k), n k This represents the total number of feasible joint events.

[0026] Optionally, forming the association cost matrix from the association costs includes the following steps:

[0027] Determine whether the calculated association cost is less than the threshold. If it is less, retain the association cost. If it is not less, calculate the association probability between the measurement of the tracked target in the current frame and the fused trajectory of the tracked target. Multiply the reciprocal of the calculated association probability with the calculated association cost and retain the product.

[0028] The association cost matrix is ​​composed of the retained association costs and the retained product, where the position of the retained product in the association cost matrix is ​​the position of the association cost that is greater than the threshold.

[0029] Optionally, the measurement and fused track of the tracked target can be correlated using a joint probability data correlation method based on the correlation cost matrix and a Hungarian matching algorithm.

[0030] The technical solution provided by this invention significantly improves target misassociation caused by errors in sensor measurements such as position and velocity by utilizing the probabilistic association results of JPDA; it fully considers the possibility of association between all sensor measurements and all targets. By combining the joint probabilistic data association algorithm with matching algorithms for position, velocity, and other information, it avoids multi-sensor target misassociation caused by sensor calibration errors, inherent performance issues, and environmental interference. After introducing the joint probabilistic data association algorithm, the association probability between misassociated measurements and tracks caused by large errors in position and velocity is usually low, thus significantly improving this type of misassociation problem in subsequent matching processes. Compared to existing technologies that use a constant velocity model to recursively obtain a one-step prediction of the target state, then obtain corresponding measurement predictions based on the observation equations of millimeter-wave radar and cameras, and finally calculate the weighted norm of the residual vector between the measurement predictions and the corresponding acquired measurements as the statistical distance for target association, and then repeatedly recombine based on this statistical distance to obtain a new association hypothesis matrix, and select the feasible association event with the smallest sum of statistical distances, the technical solution provided in this application has a simpler process and can find a good balance between accuracy and complexity, saving computing resources.

[0031] Furthermore, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target tracking method described in any of the above-mentioned embodiments.

[0032] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the target tracking method described in any one of the above claims.

[0033] These features and advantages of the present invention will be disclosed in detail in the following specific embodiments and accompanying drawings. The preferred embodiments or means of the present invention will be shown in detail in conjunction with the accompanying drawings, but are not intended to limit the technical solutions of the present invention. In addition, each of these features, elements and components appearing in the following text and drawings is a plurality of, and different symbols or numbers are used for convenience of representation, but all represent parts with the same or similar construction or function. Attached Figure Description

[0034] The present invention will be further described below with reference to the accompanying drawings:

[0035] Figure 1 This is a flowchart of an embodiment of the present invention;

[0036] Figure 2 This is a schematic diagram of the CTRV motion model in an embodiment of the present invention. Detailed Implementation

[0037] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described are intended to explain the present invention and should not be construed as limiting the invention.

[0038] The terms "an embodiment," "example," or "trademark" used in this specification refer to a particular feature, structure, or characteristic described in connection with the embodiment itself that may be included in at least one embodiment disclosed in this patent. The phrase "in an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0039] Example:

[0040] like Figure 1 As shown, this embodiment provides a target tracking method for vehicles with autonomous driving capabilities to track targets during autonomous driving. The target refers to the vehicle in front of the vehicle operating autonomous driving. The target tracking method provided in this embodiment includes the following steps iteratively:

[0041] Various sensors installed on the vehicle perceive the tracked target and acquire measurements of the target. Vehicles with autonomous driving capabilities are equipped with multiple sensors around the vehicle, such as cameras, LiDAR, and millimeter-wave radar, to perceive moving and static targets in autonomous driving scenarios. The data obtained by these sensors in measuring the tracked target constitutes the respective sensor's measurement of the tracked target.

[0042] Calculate the association cost between the measured values ​​of the tracked target and the predicted current state of the tracked target. In this step, the state of the target is first predicted using the CTRV kinematic model on the historical track, and the current state of the target is recursively obtained based on the target state estimate from the previous moment. The CTRV kinematic model is used to predict the state of the track, including longitudinal position x, lateral position y, longitudinal velocity Vx, lateral velocity Vy, yaw angle θ, and turning rate θ′. For example, if the state vector of the track in the previous second [xy Vy] is known... x V y θ θ'] T Now we need to calculate the state vector of the trajectory at the current moment, which can be done using the CTRV kinematic model. The CTRV kinematic model assumes that the object moves at a constant speed and a constant turning rate, such as... Figure 2 As shown.

[0043] Figure 2 This represents the state change of the target vehicle from time K to time K+1, according to the integral formula:

[0044]

[0045] Assuming the time interval between time k and time k+1 is Δt, expanding the above equation, we get:

[0046]

[0047] Since the CTRV kinematic model assumes a constant velocity, the velocity v is a constant independent of time t and can be placed outside the integral. Secondly, expressing cos(θ(t)) and sin(θ(t)) as functions of t, the final formula to be integrated is as follows:

[0048]

[0049] Integrating this equation yields the formula for predicting the current state of the tracked target:

[0050]

[0051] Where k is time k, k+1 is the next time k, and S k+1 S is the predicted value of the tracked target at time k in the next time step. k Let v be the state value of the tracked target at time k. k Let θ be the velocity of the tracked target. k Let θ' be the yaw angle of the tracked target at time k. k Let Δt be the turning speed of the tracked target at time k, and let Δt be the time interval between time k and time k+1.

[0052] The CTRV kinematic model can predict the current state based on the historical state of the track, thereby enabling better data correlation among the measurements of various sensors.

[0053] Then, the measurements of the tracked target are filtered according to the following sub-steps:

[0054] Traverse the merged track of the tracked target;

[0055] Query whether there are any associated measurements in the previous frame of the fused track of the tracked target. If yes, proceed to the next step. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target.

[0056] In the current frame of the fused track of the tracked target, query whether there is a measurement with the same sensor ID as an already associated measurement. If yes, calculate the association cost between the measurement of the tracked target and the predicted current state of the tracked target. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target. The formula for calculating the association probability is:

[0057]

[0058] Where, β jt (k) represents the association probability between the j-th measurement and the t-th target, P{δ i (k)|Z k} represents the i-th joint event δ at time k. i The conditional probability of (k), where nk is the total number of feasible joint events.

[0059] After screening, the association cost between the measured values ​​of the screened tracked targets and the predicted current state values ​​of the tracked targets is calculated according to the following formula:

[0060] Cost l =1.0×|x l -x k |+0.8×|y l -y k |+0.8×|Vx l -Vx k |+0.8×|Vy l -Vy k |

[0061] Cost v =1.0×|x v -x k |+1.4×|y v -y k |+0.2×|Vx v -Vx k|+0.2×|Vy v -Vy k |

[0062] Cost r =1.0×|x r -x k |+0.7×|y r -y k |+1.0×|Vx r -Vx k |+1.0×|Vy r -Vy k |

[0063] Where Cost is the association cost, x is the longitudinal position of the tracked target in the global coordinate system, vy is the lateral position of the tracked target in the global coordinate system, Vx is the longitudinal velocity of the tracked target in the global coordinate system, and Vy is the longitudinal velocity of the tracked target in the global coordinate system. The subscript k indicates that the corresponding parameter at time k is the true value; the subscript l indicates that the corresponding parameter at time k is the measurement of the LiDAR; the subscript v indicates that the corresponding parameter at time k is the measurement of the visual sensor; and the subscript r indicates that the corresponding parameter at time k is the measurement of the millimeter-wave radar. In this embodiment, for the LiDAR, the longitudinal position, lateral position, longitudinal velocity, and lateral velocity have high measurement accuracy and performance, therefore they all have high weights. Since the camera's ability to estimate target velocity is relatively poor, the longitudinal velocity and lateral velocity are given low weights of 0.2. The millimeter-wave radar's ability to estimate lateral position is slightly less than its ability to estimate longitudinal position, therefore its corresponding weight is 0.7.

[0064] In existing technologies, the correlation cost only uses longitudinal and lateral positions as metrics. However, this embodiment uses longitudinal position, lateral position, longitudinal velocity, and lateral velocity as metrics for the correlation cost, and assigns different weights according to the performance of different sensors, thus avoiding the impact of a single observation error on the accuracy of measurement and track correlation.

[0065] The association cost matrix is ​​formed from the association costs using the following steps:

[0066] Determine whether the calculated association cost is less than the threshold. If it is less, retain the association cost. If it is not less, calculate the association probability between the measurement of the tracked target in the current frame and the fused trajectory of the tracked target. Multiply the reciprocal of the calculated association probability with the calculated association cost and retain the product.

[0067] The association cost matrix is ​​composed of the retained association costs and the retained product, where the position of the retained product in the association cost matrix is ​​the position of the association cost that is greater than the threshold.

[0068] Then, using a joint probabilistic data association method, the Hungarian matching algorithm is employed to associate the measurements and fused tracks of the tracked target based on the association cost matrix. Measurements and tracks with lower costs in the association cost matrix are more likely to form associations, while those with higher costs are less likely to form associations. Simultaneously, the Hungarian matching algorithm ensures that as many measurement-track associations as possible are found. The Hungarian matching algorithm is existing technology in this field and will not be elaborated upon here. After each sensor's independent association and tracking, the ID of the same target is generally the same in consecutive frames. Therefore, associating measurements and tracks based on their IDs first is highly effective and can largely eliminate interference from other measurements.

[0069] The technical solution of this embodiment assumes, through the association assumption of the joint data probability method, that the multi-sensor fused track may or may not be associated with all sensor measurements, and the probability of association or non-association is given by the association probability. For two objects with a high association probability, the original association cost is appropriately reduced, thereby achieving a higher probability of association between the two objects during the Hungarian matching process. Conversely, for two objects with a low association probability, it is almost impossible to associate them during the Hungarian matching process. This improved scheme has significant anti-interference capabilities against the problem of false association caused by errors in sensor position and velocity measurements, resulting in excessively high or low association costs. It greatly improves the accuracy of association between multi-sensor measurement targets, providing more accurate measurement information for the fused track in the subsequent Kalman filtering process.

[0070] Finally, the state of the tracked target is updated through filtering, ending target tracking for the current frame. Aluminum foil updating is a standard filtering method in this field and will not be elaborated upon here. Target tracking for the next frame begins with the input of the sensor measurement signal. After ending tracking for the current frame, multi-sensor tracking begins again with the input of the sensor measurement signal for the next frame, and this process is repeated iteratively to track the target.

[0071] The technical solution provided in this embodiment combines a joint probabilistic data association algorithm with a matching algorithm for position, velocity, and other information, avoiding multi-sensor target misassociation caused by sensor calibration errors, inherent performance issues, and environmental interference. After introducing the joint probabilistic data association algorithm, the association probability between misassociated measurements and tracks caused by large errors in position and velocity is typically low, thus significantly improving this problem in subsequent matching processes. Compared to existing technologies that use a constant velocity model to recursively obtain a one-step prediction of the target state, then obtain corresponding measurement predictions based on the observation equations of millimeter-wave radar and cameras, and finally calculate the weighted norm of the residual vector between the measurement predictions and the corresponding acquired measurements as the statistical distance for target association, and then repeatedly recombine based on this statistical distance to obtain a new association hypothesis matrix, selecting the feasible association event with the smallest sum of statistical distances, the technical solution provided in this application has a simpler process and achieves a good balance between accuracy and complexity, saving computational resources.

[0072] Meanwhile, this embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it causes the processor to perform the steps of the target tracking method described above. The steps of the target tracking method here can be steps from the memory analysis methods of the various embodiments described above.

[0073] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. Accordingly, the computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can implement the methods of any of the above embodiments. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0074] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art should understand that the present invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments above. Any modifications that do not depart from the functional and structural principles of the present invention will be included within the scope of the claims.

Claims

1. A target tracking method for tracking targets in a vehicle with autonomous driving capabilities during autonomous driving, characterized in that, The target tracking method includes the following steps in a loop: Various sensors installed on the vehicle perceive the tracked target and acquire measurements of the tracked target; Calculate the association cost between the measurement of the tracked target and the predicted current state of the tracked target; The correlation cost is used to form a correlation cost matrix, and the measurement and fused trajectory of the tracked target are correlated based on the correlation cost matrix; The state of the tracked target is updated by filtering, target tracking in the current frame ends, and target tracking begins in the next frame; Before calculating the association cost between the measurement of the tracked target and the predicted current state of the tracked target, the measurement of the tracked target is screened. The screening process for calculating the association cost between the selected measurement of the tracked target and the predicted current state of the tracked target includes the following steps: Traverse the merged track of the tracked target; Query whether there are any associated measurements in the previous frame of the fused track of the tracked target. If yes, proceed to the next step. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target. In the current frame of the fused track of the tracked target, query whether there is a measurement with the same sensor ID as the already associated measurement. If yes, calculate the association cost between the measurement of the tracked target and the current state prediction value of the tracked target. If no, calculate the association probability between the measurement of the tracked target in the current frame and the fused track of the tracked target. The formation of the association cost matrix from the association cost includes the following steps: Determine whether the calculated association cost is less than the threshold. If it is less, retain the association cost. If it is not less, calculate the association probability between the measurement of the tracked target in the current frame and the fused trajectory of the tracked target. Multiply the reciprocal of the calculated association probability with the calculated association cost and retain the product. The association cost matrix is ​​composed of the retained association costs and the retained product, where the position of the retained product in the association cost matrix is ​​the position of the association cost that is greater than the threshold.

2. The target tracking method according to claim 1, characterized in that, The current state prediction of the tracked target is obtained using the following formula: in, k For a moment, k +1 is k The next moment of the moment, For the tracked target in k The predicted value at the next time step. For the tracked target in k The state value at time 10:00 The speed of the tracked target, For the tracked target in k Yaw angle at any moment For the tracked target in k Turning speed at any moment for k Time and k +1 is the time interval.

3. The target tracking method according to claim 1, characterized in that, The association cost between the measured value of the tracked target and the predicted value of the current state of the tracked target is calculated using the following formula: in, Cost For associated costs, x This represents the vertical position of the tracked target in the global coordinate system. y This represents the horizontal position of the tracked target in the global coordinate system. Vx The longitudinal velocity of the tracked target in the global coordinate system. Vy The longitudinal velocity of the tracked target in the global coordinate system; the subscript is... k Indicates in k The parameter at that time is the true value, and the subscript is... l Indicates in k The corresponding parameters at that time are the measurements from the lidar; the subscript is... v Indicates in k The corresponding parameter at that moment is the measurement of the vision sensor; the subscript is... r Indicates in k The corresponding parameters at that time are measured by millimeter-wave radar.

4. The target tracking method according to claim 1, characterized in that, The correlation probability between the measurement of the tracked target in the current frame and the fused trajectory of the tracked target is calculated using the following formula: in, Let be the probability of association between the j-th measurement and the t-th target. For the i-th joint event at time k The conditional probability, n k This represents the total number of feasible joint events.

5. The target tracking method according to any one of claims 1 to 4, characterized in that, The measurement and fused track of the tracked target are correlated by the joint probability data association method and the Hungarian matching algorithm based on the association cost matrix.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the target tracking method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the target tracking method according to any one of claims 1 to 5.

8. A motor vehicle, characterized in that, The motor vehicle has an automatic driving function. When the motor vehicle operates the automatic driving function, it performs target tracking using the target tracking method described in any one of claims 1 to 5. Or the motor vehicle may have the computer equipment as described in claim 6; Alternatively, the motor vehicle may have a computer-readable storage medium as described in claim 7, wherein the computer program, when executed by a processor, implements the target tracking method as described in any one of claims 1 to 5.

Images