Target tracking method, device, apparatus and storage medium

Abstract

The application discloses a target tracking method, device and equipment and a storage medium. The target tracking method comprises the following steps: acquiring a plurality of first observation state matrices of a current moment based on point cloud data of the current moment detected by a detection device, wherein the first observation state matrices comprise first state parameters of corresponding target detection boxes; determining speed information of the target detection boxes corresponding to each first observation state matrix of the current moment respectively, and performing dimension expansion on the corresponding first observation state matrix of the current moment based on the speed information to obtain a plurality of second observation state matrices of the current moment, wherein the second observation state matrices comprise second state parameters of the corresponding target detection boxes, and the second state parameters comprise the first state parameters and the speed information; and updating the state of the target of the current moment based on the plurality of second observation state matrices of the current moment. In the foregoing manner, the accuracy of target tracking can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to a target tracking method, apparatus, device, and storage medium. Background Technology

[0002] In the field of intelligent driving technology, it is usually necessary to detect the state parameters of targets around the vehicle and track their motion state and trajectory based on these parameters in order to plan the vehicle's driving path.

[0003] However, low target tracking accuracy can affect the accuracy of the vehicle's path planning, thereby impacting driving safety and potentially leading to accidents. Therefore, accurately tracking targets around the vehicle has become a pressing technical challenge. Summary of the Invention

[0004] The main technical problem solved by this invention is to provide a target tracking method, apparatus, device, and computer-readable storage medium that can improve the accuracy of target tracking.

[0005] To address the aforementioned technical problems, this application provides a target tracking method, comprising: acquiring several first observation state matrices at the current moment based on point cloud data detected by a detection device, wherein each first observation state matrix includes first state parameters of corresponding target detection boxes; determining the velocity information of the target detection boxes corresponding to each first observation state matrix at the current moment, and expanding the dimension of the corresponding first observation state matrices at the current moment based on the velocity information to obtain several second observation state matrices at the current moment, wherein each second observation state matrix includes second state parameters of the corresponding target detection boxes, and the second state parameters include first state parameters and velocity information; and updating the current state of the target based on the several second observation state matrices at the current moment.

[0006] Optionally, the velocity information of the target detection box corresponding to each first observation state matrix at the current time is determined, including: determining at least one first observation state matrix at the previous time; for each first observation state matrix at the current time, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is determined based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time and the position reference data of the target detection box corresponding to at least one first observation state matrix at the previous time.

[0007] Optionally, the position reference data is point cloud data within the target detection box. Based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time and the position reference data of the target detection boxes corresponding to at least one first observation state matrix at the previous time, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is determined, including: performing point cloud registration processing on the point cloud data within the target detection box corresponding to the first observation state matrix at the current time and the point cloud data within the target detection boxes corresponding to at least one first observation state matrix at the previous time to obtain at least one translation parameter; differentiating the at least one translation parameter to obtain at least one first velocity; and combining the at least one first velocity to obtain the velocity information of the target detection box corresponding to the first observation state matrix at the current time.

[0008] Optionally, the position reference data is the position information of the target detection box. Based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time and the position reference data of the target detection boxes corresponding to at least one first observation state matrix at the previous time, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is determined, including: based on the position information of the target detection box corresponding to the first observation state matrix at the current time and the position information of the target detection box corresponding to at least one first observation state matrix at the previous time, the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time are determined; based on the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time, at least one second velocity is obtained; and by combining at least one second velocity, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is obtained.

[0009] Optionally, before determining the velocity information of the target detection box corresponding to each first observation state matrix at the current time, the method further includes: determining the first association probability corresponding to each first observation state matrix at the previous time, wherein the first association probability is used to represent the probability that the state parameter of the first observation state matrix is ​​the state parameter of the target; determining at least one first observation state matrix at the previous time, including: determining at least one first observation state matrix at the previous time based on the first association probability corresponding to each first observation state matrix at the previous time.

[0010] Optionally, based on the first association probabilities corresponding to each first observation state matrix at the previous time step, at least one first observation state matrix at the previous time step is determined, including: taking the first observation state matrix with the largest first association probability at the previous time step as at least one first observation state matrix at the previous time step.

[0011] Optionally, the target's current state is updated based on several second observation state matrices at the current time, including: performing Kalman state updates based on each of the second observation state matrices at the current time to obtain a first estimated state matrix corresponding to each of the second observation state matrices at the current time; weighting each of the first estimated state matrices at the current time based on the first association probabilities corresponding to each of the second observation state matrices at the current time to obtain a second estimated state matrix at the current time; using the second estimated state matrix at the current time as the target's current state matrix; or, determining a second association probability at the current time, where the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the target's state parameters; multiplying the second association probability at the current time with the predicted state matrix at the current time to obtain a third estimated state matrix at the current time; and using the sum of the second estimated state matrix at the current time and the third estimated state matrix at the current time as the target's current state matrix.

[0012] Optionally, based on the point cloud data detected by the detection device at the current moment, several first observation state matrices at the current moment are obtained, including: based on the point cloud data at the current moment, determining several original observation state matrices at the current moment; and selecting the original observation state matrices associated with the target from the several original observation state matrices at the current moment as several first observation state matrices at the current moment.

[0013] Optionally, the method further includes: determining the second association probability at the current time and the second association probability at the previous time, wherein the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target; in response to the second association probability at the current time being equal to a preset probability threshold, using the predicted state matrix at the current time as the state matrix of the target at the current time; in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being equal to the preset probability threshold, updating the state of the target at the current time based on several first observation state matrices at the current time; in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being less than the preset probability threshold, performing the determination of the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time and subsequent steps thereof.

[0014] To address the aforementioned technical problems, another technical solution adopted in this application is: providing a target tracking device, comprising: an acquisition module, configured to acquire several first observation state matrices at the current moment based on point cloud data detected by a detection device, wherein the first observation state matrices include first state parameters of corresponding target detection boxes; a velocity expansion module, configured to determine the velocity information of the target detection boxes corresponding to each first observation state matrix at the current moment, and expand the dimension of the corresponding first observation state matrices at the current moment based on the velocity information to obtain several second observation state matrices at the current moment, wherein the second observation state matrices include second state parameters of the corresponding target detection boxes, and the second state parameters include first state parameters and velocity information; and a state update module, configured to update the state of the target at the current moment based on the several second observation state matrices at the current moment.

[0015] Optionally, the velocity expansion module is used to determine at least one first observation state matrix of the previous time step; for each first observation state matrix of the current time step, based on the position reference data of the target detection box corresponding to the first observation state matrix of the current time step and the position reference data of the target detection box corresponding to at least one first observation state matrix of the previous time step, the velocity information of the target detection box corresponding to the first observation state matrix of the current time step is determined.

[0016] Optionally, the position reference data is the point cloud data within the target detection box. The velocity dimension expansion module is used to perform point cloud registration processing on the point cloud data within the target detection box corresponding to the first observation state matrix at the current time and the point cloud data within the target detection box corresponding to at least one first observation state matrix at the previous time, respectively, to obtain at least one translation parameter; to differentiate the at least one translation parameter, at least one first velocity is obtained; and to combine the at least one first velocity, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is obtained.

[0017] Optionally, the position reference data is the position information of the target detection box. The velocity expansion module is used to determine the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time, based on the position information of the target detection box corresponding to the first observation state matrix at the current time and the position information of the target detection box corresponding to at least one first observation state matrix at the previous time. Based on the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time, at least one second velocity is obtained. The velocity information of the target detection box corresponding to the first observation state matrix at the current time is obtained by combining the at least one second velocity.

[0018] Optionally, the target tracking device further includes an association probability determination module. Before the velocity expansion module determines the velocity information of the target detection box corresponding to each first observation state matrix at the current time, the association probability determination module is used to determine the first association probability corresponding to each first observation state matrix at the previous time. The first association probability represents the probability that the state parameter of the first observation state matrix is ​​the state parameter of the target. The velocity expansion module is used to determine at least one first observation state matrix at the previous time based on the first association probability corresponding to each first observation state matrix at the previous time.

[0019] Optionally, the velocity expansion module is used to take the first observation state matrix with the highest first correlation probability in the previous time step as at least one first observation state matrix in the previous time step.

[0020] Optionally, the state update module is used to perform Kalman state updates based on each of the second observed state matrices at the current time to obtain the first estimated state matrix corresponding to each of the second observed state matrices at the current time; based on the first association probability corresponding to each of the second observed state matrices at the current time, weighting each of the first estimated state matrices at the current time to obtain the second estimated state matrix at the current time; using the second estimated state matrix at the current time as the target's state matrix at the current time; or, determining the second association probability at the current time, where the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the target's state parameters; multiplying the second association probability at the current time with the predicted state matrix at the current time to obtain the third estimated state matrix at the current time; and using the sum of the second estimated state matrix at the current time and the third estimated state matrix at the current time as the target's state matrix at the current time.

[0021] Optionally, the acquisition module is used to determine several original observation state matrices at the current time based on the point cloud data at the current time; and to select the original observation state matrices associated with the target from the several original observation state matrices at the current time as several first observation state matrices at the current time.

[0022] Optionally, the association probability determination module is further configured to determine the second association probability at the current time and the second association probability at the previous time, where the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target. The state update module is configured to, in response to the second association probability at the current time being equal to a preset probability threshold, use the predicted state matrix at the current time as the target's current state matrix; in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being equal to the preset probability threshold, update the target's current state based on several first observed state matrices at the current time; and in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being less than the preset probability threshold, execute the determination of the velocity information of the target detection boxes corresponding to each first observed state matrix at the current time and subsequent steps.

[0023] To solve the above-mentioned technical problems, another technical solution adopted in this application is: to provide an electronic device, including a memory and a processor coupled to each other, wherein the memory stores program instructions; and the processor is used to execute the program instructions stored in the memory to implement the above-mentioned target tracking method.

[0024] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a computer-readable storage medium for storing program instructions that can be executed by a processor to implement the above-mentioned target tracking method.

[0025] The above scheme obtains several first observation state matrices based on point cloud data detected by a detection device at the current moment. It then determines the velocity information of the target detection boxes corresponding to each first observation state matrix at the current moment, and expands the dimension of the corresponding first observation state matrices based on the velocity information to obtain several second observation state matrices at the current moment. Finally, it updates the target's current state based on these second observation state matrices. The first observation state matrix includes the first state parameters of the corresponding target detection boxes, and the second observation state matrix includes the second state parameters of the corresponding target detection boxes. The second state parameters include the first state parameters and the velocity information. Since the detection device cannot directly detect the target's velocity information, the several first observation state matrices at the current moment lack velocity information, resulting in low accuracy for updating the target's current state directly based on these first observation state matrices. Therefore, updating the target's current state based on the velocity-expanded second observation state matrices at the current moment improves the accuracy of target tracking. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an embodiment of the target tracking method provided in this application;

[0027] Figure 2 This is a flowchart illustrating another embodiment of the target tracking method provided in this application;

[0028] Figure 3 This is a flowchart illustrating an embodiment of the speed information determination method provided in this application;

[0029] Figure 4 This is a flowchart illustrating an embodiment of the speed information determination method provided in this application;

[0030] Figure 5 This is a flowchart illustrating yet another embodiment of the target tracking method provided in this application;

[0031] Figure 6 This is a schematic diagram of the framework of an embodiment of the target tracking device provided in this application;

[0032] Figure 7 This is a schematic diagram of the framework of an embodiment of the electronic device provided in this application;

[0033] Figure 8 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium provided in this application. Detailed Implementation

[0034] To make the purpose, technical solution and effects of this application clearer and more explicit, the following describes this application in further detail with reference to the accompanying drawings and embodiments.

[0035] In the following description, specific details such as particular system architectures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0036] In this article, the term "several" means at least one, and the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0037] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the target tracking method provided in this application. It should be noted that if substantially the same result is achieved, the method of this invention is not necessarily identical. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown, the method includes the following steps:

[0038] S11: Based on the point cloud data obtained by the detection device at the current moment, obtain several first observation state matrices at the current moment.

[0039] In this embodiment, the point cloud data detected by the detection device at the current moment includes the position information of all points on all detected targets at the current moment. For example, the detection device may be a lidar, a camera, etc., and this embodiment does not specifically limit it.

[0040] By performing target detection on the point cloud data obtained by the detection device at the current moment, several original observation state matrices at the current moment are obtained, and several first observation state matrices at the current moment are determined based on the several original observation state matrices at the current moment. For example, the target detection can be three-dimensional target detection or two-dimensional target detection.

[0041] Each first observation state matrix includes the first state parameters of the corresponding target detection box. It should be noted that, since the detection device cannot directly detect the target's velocity information, the velocity information of the target detection box is not included in the first state parameters. For example, the first state parameters may include the target detection box's position information, size information, heading angle, and confidence score, among other state parameters. Taking 3D target detection as an example, the target detection box's position information is the coordinates of its center point in the vehicle coordinate system; the coordinates of the center point on the x, y, and z axes of the vehicle coordinate system can be represented by x, y, and z, respectively. The target detection box's size information includes its length, width, and height, which can be represented by l, w, and h, respectively. The target detection box's heading angle and confidence score can be represented by θ and s, respectively.

[0042] In one embodiment, original observation state matrices associated with the target are selected from a plurality of original observation state matrices at the current time, and these are used as a plurality of first observation state matrices at the current time. The state parameters in each of the first observation state matrices associated with the target may all be state parameters of the target at the current time, and the probability that the state parameters in each of the first observation state matrices associated with the target are state parameters of the target is not the same.

[0043] In another embodiment, several original observation state matrices at the current time are directly used as several first observation state matrices at the current time.

[0044] S12: Determine the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time.

[0045] In step S12, at least one first observation state matrix from the previous time step can be determined first. Then, for each first observation state matrix at the current time step, based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time step and the position reference data of the target detection boxes corresponding to at least one first observation state matrix from the previous time step, the velocity information of the target detection box corresponding to the first observation state matrix at the current time step is determined. For example, the position reference data of the target detection box can be point cloud data within the target detection box, or the position reference data of the target detection box can be the position information of the target detection box.

[0046] S13: Based on the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time, the dimensions of the corresponding first observation state matrix at the current time are expanded to obtain several second observation state matrices at the current time.

[0047] The second observation state matrix includes second state parameters of the corresponding target detection boxes. These second state parameters include first state parameters and the velocity information of the target detection boxes. For example, the velocity information of the target detection boxes includes velocities in the x-axis, y-axis, and z-axis directions of the vehicle coordinate system, for example, represented by v0. x v y and v z The first observation state matrix at the current moment can be represented as: [x,y,z,θ,l,w,h,s]. T The second observation state matrix obtained by expanding the dimension of the first observation state matrix can be represented as: [x,y,z,θ,l,w,h,s,v] x ,v y ,v z ] T .

[0048] S14: Update the target's current state based on several second observation state matrices at the current moment.

[0049] In this embodiment, several first observation state matrices are obtained based on the point cloud data detected by the detection device at the current moment. The velocity information of the target detection box corresponding to each first observation state matrix at the current moment is determined, and the corresponding first observation state matrix at the current moment is expanded in dimension based on the velocity information to obtain several second observation state matrices at the current moment. The target's current state is updated based on these second observation state matrices. The first observation state matrix includes the first state parameters of the corresponding target detection box, and the second observation state matrix includes the second state parameters of the corresponding target detection box. The second state parameters include the first state parameters and the velocity information. Since the detection device cannot directly detect the target's velocity information, the several first observation state matrices at the current moment lack velocity information, resulting in low accuracy of the target's current state updated directly based on these first observation state matrices. Therefore, updating the target's current state based on the velocity-expanded several second observation state matrices at the current moment can improve the accuracy of target tracking.

[0050] Please see Figure 2 , Figure 2 This is a flowchart illustrating another embodiment of the target tracking method provided in this application. Figure 2 As shown, the method includes the following steps:

[0051] S21: Based on the point cloud data obtained by the detection device at the current moment, obtain several first observation state matrices at the current moment.

[0052] In this embodiment, each first observation state matrix includes first state parameters of the corresponding target detection box, but the first state parameters do not include the velocity information of the target detection box. Step S21 includes: determining several original observation state matrices at the current time based on the point cloud data detected by the detection device; and selecting the original observation state matrices associated with the target from the several original observation state matrices at the current time as several first observation state matrices at the current time. The state parameters of each first observation state matrix at the current time may all be the state parameters of the target at the current time, and the probability that the state parameters of each first observation state matrix at the current time are the state parameters of the target at the current time is not the same.

[0053] In one embodiment, determining several original observation state matrices at the current moment based on the point cloud data detected by the detection device includes: performing three-dimensional target detection on the point cloud data detected by the detection device at the current moment to obtain several original observation state matrices at the current moment. The state parameters in the several original observation state matrices at the current moment may be the state parameters of the target at the current moment, the state parameters of other targets at the current moment, or noise data caused by interference signals.

[0054] In one embodiment, selecting a plurality of first observation state matrices associated with the target from a plurality of original observation state matrices at the current moment includes the following sub-steps:

[0055] Sub-step one involves performing Kalman prediction based on the target's state matrix from the previous time step to obtain the target's predicted state matrix at the current time step, and determining the prediction covariance matrix corresponding to the Kalman prediction at the current time step. The prediction covariance matrix represents the error between the predicted state matrix at the current time step and the target's true state matrix at the current time step.

[0056] In a specific application, assuming the target is moving at a constant speed, the predicted state matrix of the target at the current moment can be obtained using the following formula:

[0057]

[0058] In formula (1), k represents the current time, and k-1 represents the previous time. Let F represent the predicted state matrix of the target at the current moment, and let F represent the state transition matrix. This represents the state matrix of the target at the previous moment.

[0059] In a specific application, the prediction covariance matrix corresponding to the Kalman prediction at the current time can be determined using the following formula:

[0060] P k|k-1 =F*P k-1|k-1 *F T +Q (2)

[0061] In formula (2), P k|k-1 Let P represent the prediction covariance matrix at the current time, F represent the state transition matrix, and P represent the prediction covariance matrix at the current time. k-1|k-1 Let represent the covariance matrix of the target at the previous time step, and Q represent the prediction noise matrix.

[0062] Sub-step two: Determine the predicted observation state matrix at the current time based on the predicted state matrix at the current time, and determine the innovation covariance matrix at the current time based on the predicted covariance matrix at the current time.

[0063] In a specific application, the predicted observation state matrix at the current moment is determined using the following formula:

[0064]

[0065] In formula (3), H represents the predicted observation state matrix at the current moment, and H represents the measurement matrix. This represents the predicted state matrix at the current moment.

[0066] In a specific application, the current information covariance matrix is ​​determined using the following formula:

[0067] S k =H*P k|k-1 *H T +R (4)

[0068] In formula (4), S k Let H represent the information covariance matrix at the current moment, and let P represent the measurement matrix. k|k-1 Let R represent the prediction covariance matrix at the current moment, and let R represent the measurement noise matrix.

[0069] Sub-step three: Based on the predicted observation state matrix and the information covariance matrix at the current time, select several first observation state matrices associated with the target from several original observation state matrices at the current time.

[0070] In a specific application, several first observation state matrices are obtained by filtering using the following formula:

[0071]

[0072] In formula (5), y represents the original observation state matrix among several original observation state matrices. S represents the predicted observation state matrix at the current moment. k This represents the new information covariance matrix at the current moment. For effective gate parameters, The value is determined by the dimension of the state variable and the chi-square distribution table under the query tracking gate rule. When the original observation state matrix satisfies the condition of formula (5), it is determined that the original observation state matrix is ​​associated with the target. When the original observation state matrix does not satisfy the condition of formula (5), it is determined that the original observation state matrix is ​​not associated with the target.

[0073] The selected first observation state matrices at the current time can be represented by the following expression:

[0074]

[0075] In expression (6), Let y represent the set of the first observed state matrices at the current moment. k (i) represents the i-th first observation state matrix at the current time, i∈{1,2,…m} k}, m k This represents the number of first observation state matrices in the set of first observation state matrices at the current moment.

[0076] In this embodiment, by selecting several first observation state matrices associated with the target from several original observation state matrices at the current moment, original observation state matrices that are not associated with the target can be filtered out, thereby further improving the accuracy of subsequent determination of velocity parameters.

[0077] S22: Determine at least one first observation state matrix from the previous time step.

[0078] In this embodiment, at least one first observation state matrix at the previous time step is a portion of the plurality of first observation state matrices at the previous time step, or, at least one first observation state matrix at the previous time step is the plurality of first observation state matrices at the previous time step. The plurality of first observation state matrices at the previous time step can be obtained based on the point cloud data at the previous time step detected by the detection device. For relevant details, please refer to the aforementioned steps S11 or S21, which will not be repeated here.

[0079] In one embodiment, since each of the first observation state matrices at the previous time step is associated with the state of the target at the previous time step, in order to improve the accuracy of determining the velocity information, several first observation state matrices at the previous time step are used as at least one first observation state matrix at the previous time step.

[0080] In another embodiment, to reduce the computational load of determining velocity information, a subset of first observation state matrices from the previous time step can be selected for velocity information calculation. Specifically, a first association probability corresponding to each first observation state matrix at the previous time step is determined; based on the first association probabilities corresponding to each first observation state matrix at the previous time step, at least one first observation state matrix at the previous time step is determined. The first association probability represents the probability that the state parameters of the first observation state matrix are the state parameters of the target.

[0081] In a specific application, the first correlation probability of each first observation state matrix at the previous time step can be calculated using the following formula:

[0082]

[0083] In formula (7), β k-1 (i) represents the first correlation probability of the first observed state matrix at the previous time step i, where i is an integer greater than 0, P G P represents the probability of falling into the gate. D p represents the probability of being detected by the sensor. k-1 (i) represents the likelihood value corresponding to the i-th first observation state matrix at the previous time step, ρ k-1 This represents clutter density.

[0084] Where, p k-1(i) It can be calculated using the following formula:

[0085]

[0086] In formula (8), y k-1 (i) represents the first observation state matrix at the i-th time step in the previous time step. S represents the predicted observed state matrix at the previous time step. k-1 Let represent the information covariance matrix of the previous time step.

[0087] δ k-1 The following formula can be used to calculate it:

[0088] δ k-1 =1-Λ k-1 (9)

[0089] In formula (9), Λ k-1 Indicates likelihood ratio. Λ k-1 The following formula can be used to calculate it:

[0090]

[0091] In formula (10), m k-1 This represents the number of first observation state matrices in the set of first observation state matrices at the previous time step.

[0092] In one example, the first observation state matrix whose first association probability at the previous time step is greater than a set threshold is used as at least one first observation state matrix at the previous time step. Exemplarily, the set threshold can be set according to actual needs.

[0093] In another example, the first association probabilities corresponding to each first observation state matrix at the previous time step are sorted, and the first observation state matrix whose first association probability at the previous time step is located in a preset order is taken as at least one first observation state matrix at the previous time step. For example, the preset order can be set according to actual needs, such as the first 3 positions, the first 2 positions, etc.

[0094] In another example, since the state parameters of the first observation state matrix with the highest first correlation probability are most likely to be the target's state parameters, the first observation state matrix with the highest first correlation probability at the previous time step can be determined first. Then, this first observation state matrix with the highest first correlation probability at the previous time step can be used as at least one first observation state matrix at the previous time step. By using the first observation state matrix with the highest first correlation probability at the previous time step as at least one first observation state matrix at the previous time step, the computational load for determining velocity information can be further reduced, while also ensuring the accuracy of the determined velocity information, thereby further improving the accuracy of target tracking.

[0095] S23: For each first observation state matrix at the current time, based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time and the position reference data of the target detection box corresponding to at least one first observation state matrix at the previous time, determine the velocity information of the target detection box corresponding to the first observation state matrix at the current time.

[0096] In one embodiment, the position reference data of the target detection box can be point cloud data within the target detection box, which includes the coordinates of all points on the target within the target detection box. For each first observation state matrix at the current time, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is determined based on the point cloud data within the target detection box corresponding to the first observation state matrix at the current time and the point cloud data within the target detection boxes corresponding to at least one first observation state matrix at the previous time.

[0097] Specifically, please refer to Figure 3 , Figure 3 This is a flowchart illustrating an embodiment of the speed information determination method provided in this application, as shown below. Figure 3 As shown, the method includes the following steps:

[0098] S301, perform point cloud registration processing on the point cloud data in the target detection box corresponding to the first observation state matrix at the current time and the point cloud data in the target detection box corresponding to at least one first observation state matrix at the previous time, respectively, to obtain at least one translation parameter.

[0099] The number of at least one translation parameter corresponding to the first observation state matrix at the current time is the same as the number of at least one first observation state matrix at the previous time.

[0100] In this embodiment, for each first observation state matrix at the current time, point cloud registration is performed on the point cloud data within the target detection box corresponding to the first observation state matrix at the current time and the point cloud data within the target detection box corresponding to at least one first observation state matrix at the previous time. This yields at least one target pose transformation matrix corresponding to each first observation state matrix at the current time. Furthermore, based on the at least one target pose transformation matrix corresponding to each first observation state matrix at the current time, at least one translation parameter corresponding to each first observation state matrix at the current time is obtained. The target pose transformation matrix includes a rotation matrix and a translation matrix, and the translation parameter is the translation matrix within the target pose transformation matrix.

[0101] In this embodiment, point cloud registration can be achieved using traditional ICP (Iterative Closest Point) algorithm, NDT (Normal Distributions Transform) algorithm, GICP (Generalized Iterative Closest Point) algorithm, VGICP (Voxelized Generalized Iterative Closest Point) algorithm, or other point cloud registration algorithms. It should be noted that this embodiment does not specifically limit the type of point cloud registration algorithm.

[0102] The following section uses the VGICP algorithm as an example to briefly explain the point cloud registration process.

[0103] Taking a first observation state matrix from the previous time step and a first observation state matrix from the current time step as examples, the point cloud data within the target detection box corresponding to the first observation state matrix from the previous time step and the point cloud data within the target detection box corresponding to the first observation state matrix from the current time step are respectively represented by A k-1 and B k Indicated. Among them, A k-1 ={a0,…,a M}, B k ={b0,…,b N}, A k-1 For reference point cloud, B k The point cloud to be registered. Assume A k-1 and B k All points in A follow a Gaussian distribution and are independently distributed, i.e., A k-1 and B k The points in the middle satisfy the following expression:

[0104]

[0105] In expression (11), and Let a and b represent random variables respectively. i The mean and covariance matrices, and Let b represent random variables respectively. i The mean and covariance matrix of the random variable a i and random variable b i A respectively k-1 and B k The point in the middle.

[0106] Step 1, determine A k-1and B k The corresponding nearest neighbor distance and

[0107] When considering the nearest neighbor solution, we assume that b i In target point set A k-1 The corresponding true nearest neighbor in the value is closer to a. i And its neighboring points, and the nearest neighbor distance sum is obtained using the following formula:

[0108]

[0109] In formula (12), T represents the pose transformation matrix. Represents random variable a i The covariance matrix, b j The mean of . Where, b j B k In and T*a i Points whose distance is within the nearest neighbor distance limit threshold can be represented by the following expression:

[0110] {b j |||T*a i -b j || <r} (13)

[0111] In expression (13), r represents the nearest neighbor distance limit threshold. When used between two vehicle targets, the value of r ranges from 0.2 to 0.5.

[0112] Step 2, determine the nearest neighbor distance and The corresponding Gaussian distribution.

[0113] Nearest neighbor distance and It follows a Gaussian distribution, that is, it satisfies the following expression:

[0114]

[0115] In expression (14), and Representing nearest neighbor distance and The mean and covariance matrices, and The results are obtained using the following two formulas:

[0116]

[0117]

[0118] When the transformation is correct, all nearest neighbors are connected to T*a. i The distance should approach 0, that is, the nearest neighbor distance and the sum of the nearest neighbor distances should be equal. mean It is 0.

[0119] Step 3: Elementize the points.

[0120] The following relationship exists within each voxel:

[0121]

[0122] In expression (17), N i Indicates a i The total number of nearest neighbor points. Replacing nearest neighbor search with voxelization can improve the robustness of point cloud registration algorithms.

[0123] Step 4, based on nearest neighbor distance and The corresponding Gaussian distribution is used to determine the maximum likelihood probability, thus obtaining the target pose transformation matrix T. * .

[0124] Target pose transformation matrix T * Formula (18) can be used for calculation:

[0125]

[0126] The target pose transformation matrix T is obtained * The following relationship must be satisfied:

[0127]

[0128] Where R represents the rotation matrix and P represents the translation matrix.

[0129] S302, differentiate with respect to at least one translation parameter to obtain at least one first velocity.

[0130] For example, the first velocity is a vector; by decomposing the first velocity, the velocity information v can be obtained. x v y and v z .

[0131] S303, integrate at least one first velocity to obtain the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0132] In one example, the average value of at least one first velocity corresponding to the first observation state matrix at the current moment can be determined, and the average value of at least one first velocity can be used as the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0133] In another example, at least one first velocity corresponding to the first observation state matrix at the current moment can be weighted and summed, and the velocity obtained by weighted summation can be used as the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0134] In another embodiment, the position reference data of the target detection box can be the position information of the target detection box, such as the coordinates of the center point of the target detection box or the coordinates of the vertices of the target detection box. For each first observation state matrix at the current time, the velocity information of the target detection box corresponding to the first observation state matrix at the current time is determined based on the position information of the target detection box corresponding to the first observation state matrix at the current time and the position information of the target detection boxes corresponding to at least one first observation state matrix at the previous time.

[0135] Specifically, please refer to Figure 4 , Figure 4 This is a flowchart illustrating another embodiment of the speed information determination method provided in this application, as shown below. Figure 4 As shown, the method includes the following steps:

[0136] S401, based on the position information of the target detection box corresponding to the first observation state matrix at the current time and the position information of the target detection box corresponding to at least one first observation state matrix at the previous time, determine the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time.

[0137] The number of at least one distance corresponding to the first observation state matrix at the current time is the same as the number of at least one first observation state matrix at the previous time.

[0138] S402, based on the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection box corresponding to at least one first observation state matrix at the previous time, at least one second velocity is obtained.

[0139] Specifically, the time interval between the previous moment and the current moment is determined, and the distance is divided by the time interval to obtain the second velocity corresponding to that distance.

[0140] For example, the second velocity is a vector, and the velocity information v can be obtained by decomposing the first velocity. x v y and v z .

[0141] S403, integrate at least one second velocity to obtain the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0142] In one example, the average value of at least one second velocity corresponding to the first observation state matrix at the current moment can be determined, and the average value of at least one second velocity can be used as the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0143] In another example, at least one second velocity corresponding to the first observation state matrix at the current moment can be weighted and summed, and the velocity obtained by weighted summation can be used as the velocity information of the target detection box corresponding to the first observation state matrix at the current moment.

[0144] S24: Based on the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time, expand the dimension of the corresponding first observation state matrix at the current time to obtain several second observation state matrices at the current time.

[0145] The second observation state matrix includes the second state parameters of the corresponding target detection box, and the second state parameters include the first state parameters and the velocity information of the target detection box. The details regarding dimensionality expansion can be found in step S13 above, and will not be repeated here.

[0146] S25: Update the target's current state based on several second observation state matrices at the current moment.

[0147] In this embodiment, step S25 may include the following sub-steps:

[0148] Sub-step one: Perform Kalman state updates based on each of the second observation state matrices at the current time to obtain the first estimated state matrix corresponding to each of the second observation state matrices at the current time.

[0149] In a specific application, the first estimated state matrix corresponding to each second observation state matrix at the current time is determined using the following formula:

[0150]

[0151] In formula (20), This represents the first estimated state matrix corresponding to the i-th second observation state matrix at the current time. Let K(E) represent the predicted state matrix of the target at the current time step, and let K(E) represent the expanded Kalman gain matrix. k (i,E) represents the second observation state matrix at the current time i, and H(E) represents the expanded measurement matrix. K(E) is calculated using the following formula:

[0152] K(E)=P k|k-1 *H(E)*S k -1 (E) (21)

[0153] In formula (21), P k|k-1 S represents the prediction covariance matrix at the current time. k (E) represents the expanded information matrix at the current time, S k -1 (E) represents S k The inverse matrix of (E). Where S k (E) is calculated using the following formula:

[0154] S k (E)=H(E)*P k|k-1 *H T (E)+R(E) (22)

[0155] In formula (22), R(E) represents the expanded measurement noise matrix.

[0156] In sub-step one, the covariance matrix corresponding to each second observation state matrix at the current time can also be determined simultaneously, specifically using the following formula:

[0157]

[0158] In formula (23), Let P represent the covariance matrix corresponding to the i-th second observation state matrix at the current time. k|k-1 This represents the prediction covariance matrix at the current moment.

[0159] Sub-step two: Based on the first correlation probability corresponding to each second observation state matrix at the current time, weight the first estimated state matrices at the current time to obtain the second estimated state matrix at the current time.

[0160] Sub-step three: Based on the second estimated state matrix at the current time, determine the target's state matrix at the current time.

[0161] In one embodiment, the second estimated state matrix at the current time is used as the target's state matrix at the current time. In this case, the target's state matrix at the current time can be represented by the following formula:

[0162]

[0163] In formula (24), β k (i) represents the first correlation probability corresponding to the i-th second observation state matrix at the current time. It represents the first estimated state matrix corresponding to the i-th second observation state matrix at the current time.

[0164] In another embodiment, a second association probability is determined at the current time, representing the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target. The second association probability at the current time is multiplied by the predicted state matrix at the current time to obtain a third estimated state matrix at the current time. The sum of the second estimated state matrix and the third estimated state matrix at the current time is taken as the target's state matrix at the current time. In this case, the target's state matrix at the current time can be represented by the following formula:

[0165]

[0166] In formula (25), β k (0) represents the second correlation probability at the current time. Let β represent the third estimated state matrix at the current time. k (0) and m k β k The sum of the probabilities of (i) is 1. β k (0) can be calculated using the following expression:

[0167]

[0168] In sub-step three, the covariance matrix of the target at the current time can also be determined simultaneously based on the covariance matrices corresponding to each of the second observation state matrices at the current time. The covariance matrix of the target at the current time is used to predict the prediction covariance matrix at the next time step. Specifically, the covariance matrix of the target at the current time is determined using the following formula:

[0169]

[0170] In this embodiment, the velocity information of the target detection box corresponding to the first observation state matrix at the current time can be determined based on the position reference data of the target detection box corresponding to the first observation state matrix at the current time and the position reference data of the target detection boxes corresponding to at least one first observation state matrix at the previous time. Then, based on the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time, the corresponding first observation state matrix at the current time is expanded in dimension to obtain several second observation state matrices at the current time. Since the expanded second observation state matrices at the current time include the velocity information of the corresponding target detection boxes, updating the target's current state based on these expanded second observation state matrices can improve the accuracy of target tracking.

[0171] Please see Figure 5 , Figure 5 This is a flowchart illustrating another embodiment of the target tracking method provided in this application. Figure 5As shown, the method includes the following steps:

[0172] S51: Determine the second association probability at the current time and the second association probability at the previous time.

[0173] The second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target. The relevant content for calculating the second association probability can be referred to the aforementioned formula (26) and formulas (8) to (10), which will not be repeated here.

[0174] S52: In response to the second correlation probability at the current time being equal to the preset probability threshold, the predicted state matrix at the current time is used as the target's state matrix at the current time.

[0175] If the second correlation probability at the current moment is equal to the preset probability threshold, it means that the acquired original observation state matrices are not correlated with the target (i.e., the target detection box is not within the gate). In this case, the predicted state matrix at the current moment is believed to be based on Kalman prediction. The predicted state matrix at the current moment can be referred to the aforementioned formula (1). The preset probability threshold is 1.

[0176] S53: In response to the second correlation probability at the current moment being less than the preset probability threshold, and the second correlation probability at the previous moment being equal to the preset probability threshold, update the state of the target at the current moment based on several first observation state matrices at the current moment.

[0177] If the second association probability at the current moment is less than a preset probability threshold, and the second association probability at the previous moment is equal to the preset probability threshold, it indicates that a first observation state matrix associated with the target exists at the current moment, but no first observation state matrix associated with the target exists at the previous moment. In this case, it is impossible to determine the velocity information based on the position reference information of the target detection box corresponding to the first observation state matrix at the current moment and the position reference information of the target detection box corresponding to the first observation state matrix at the previous moment.

[0178] In this embodiment, the current state of the target is updated based on several first observation state matrices at the current moment, including the following sub-steps:

[0179] Sub-step one: Perform Kalman state updates based on each of the first observation state matrices at the current time to obtain the fourth estimated state matrix corresponding to each of the first observation state matrices at the current time.

[0180] The fourth estimated state matrix corresponding to each of the first observed state matrices at the current time is determined using the following formula:

[0181]

[0182] In formula (28), This represents the fourth estimated state matrix corresponding to the i-th first observed state matrix at the current time. Let y represent the predicted state matrix of the target at the current time step, K represent the unexpanded Kalman gain matrix, and y represent the predicted state matrix of the target at the current time step. k (i) represents the first observation state matrix at the current time i, and H represents the measurement matrix without dimension expansion.

[0183] In sub-step two, the covariance matrix corresponding to each first observation state matrix at the current time can also be determined simultaneously, specifically using the following formula:

[0184]

[0185] Sub-step two: Based on the first correlation probability corresponding to each first observation state matrix at the current time, weight the fourth estimated state matrices at the current time to obtain the fifth estimated state matrix at the current time.

[0186] Sub-step three: Based on the fifth estimated state matrix at the current time, determine the target's state matrix at the current time.

[0187] In one embodiment, the fifth estimated state matrix at the current time is used as the target's state matrix at the current time.

[0188] In another embodiment, a second association probability at the current time is determined; the second association probability at the current time is multiplied by the predicted state matrix at the current time to obtain a third estimated state matrix at the current time; the sum of the fifth estimated state matrix at the current time and the third estimated state matrix at the current time is taken as the target state matrix at the current time.

[0189] The relevant content of sub-step three can be referred to the aforementioned formulas (24) and (25), and will not be repeated here.

[0190] In sub-step three, the covariance matrix of the target at the current time can also be determined based on the covariance matrices corresponding to the first observation state matrices at the current time. The covariance matrix of the target at the current time is used to predict the prediction covariance matrix at the next time. The determination of the covariance matrix of the target at the current time can be referred to the aforementioned formula (27), which will not be repeated here.

[0191] S54: In response to the fact that the second association probability at the current time is less than the preset probability threshold, and the second association probability at the previous time is less than the preset probability threshold, execute the velocity information of the target detection box corresponding to each first observation state matrix at the current time and its subsequent steps.

[0192] The relevant content of step S54 can be referred to in steps S22 to S25 above, and will not be repeated here.

[0193] Optionally, in this embodiment, the target tracking method further includes: in n consecutive observations of the target, counting the number of times the second association probability is equal to a preset probability threshold; if the number of times the second association probability is equal to the preset probability threshold is greater than or equal to a set number, it is considered that the target is no longer within the detection range of the vehicle, and the trajectory corresponding to the target can be deleted.

[0194] In this embodiment, target tracking is performed in three different ways based on the second association probability at the current time and the second association probability at the previous time. Different optimal processing can be performed for different situations, which further improves the target tracking effect.

[0195] Please see Figure 6 , Figure 6 This is a schematic diagram of a target tracking device according to an embodiment of the present application. In this embodiment, the target tracking device 60 includes: an acquisition module 61, a velocity expansion module 62, and a status update module 63.

[0196] The acquisition module 61 is used to acquire several first observation state matrices at the current moment based on the point cloud data detected by the detection device. Each first observation state matrix includes first state parameters of the corresponding target detection boxes. The velocity expansion module 62 is used to determine the velocity information of the target detection boxes corresponding to each first observation state matrix at the current moment, and expand the dimension of the corresponding first observation state matrix at the current moment based on the velocity information to obtain several second observation state matrices at the current moment. Each second observation state matrix includes second state parameters of the corresponding target detection boxes, and the second state parameters include the first state parameters and velocity information. The state update module 63 is used to update the current state of the target based on the several second observation state matrices at the current moment.

[0197] Optionally, the velocity expansion module 62 is used to determine at least one first observation state matrix of the previous time step; for each first observation state matrix of the current time step, based on the position reference data of the target detection box corresponding to the first observation state matrix of the current time step and the position reference data of the target detection box corresponding to at least one first observation state matrix of the previous time step respectively, the velocity information of the target detection box corresponding to the first observation state matrix of the current time step is determined.

[0198] Optionally, the position reference data is the point cloud data within the target detection box. The velocity expansion module 62 is used to perform point cloud registration processing on the point cloud data within the target detection box corresponding to the first observation state matrix at the current time and the point cloud data within the target detection box corresponding to at least one first observation state matrix at the previous time, respectively, to obtain at least one translation parameter; to differentiate the at least one translation parameter, respectively, to obtain at least one first velocity; and to combine the at least one first velocity to obtain the velocity information of the target detection box corresponding to the first observation state matrix at the current time.

[0199] Optionally, the position reference data is the position information of the target detection box. The velocity expansion module 62 is used to determine the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time, based on the position information of the target detection box corresponding to the first observation state matrix at the current time and the position information of the target detection box corresponding to at least one first observation state matrix at the previous time. Based on the distances between the target detection box corresponding to the first observation state matrix at the current time and the target detection boxes corresponding to at least one first observation state matrix at the previous time, at least one second velocity is obtained. The velocity information of the target detection box corresponding to the first observation state matrix at the current time is obtained by combining the at least one second velocity.

[0200] Optionally, the target tracking device 60 further includes an association probability determination module 64. Before the velocity expansion module 62 determines the velocity information of the target detection box corresponding to each first observation state matrix at the current time, the association probability determination module 64 is used to determine the first association probability corresponding to each first observation state matrix at the previous time. The first association probability is used to represent the probability that the state parameter of the first observation state matrix is ​​the state parameter of the target. The velocity expansion module 62 is used to determine at least one first observation state matrix at the previous time based on the first association probability corresponding to each first observation state matrix at the previous time.

[0201] Optionally, the velocity expansion module 62 is used to take the first observation state matrix with the highest first correlation probability in the previous time step as at least one first observation state matrix in the previous time step.

[0202] Optionally, the state update module 63 is used to perform Kalman state updates based on each of the second observed state matrices at the current time to obtain the first estimated state matrix corresponding to each of the second observed state matrices at the current time; to perform weighted processing on each of the first estimated state matrices at the current time based on the first association probability corresponding to each of the second observed state matrices at the current time to obtain the second estimated state matrix at the current time; to use the second estimated state matrix at the current time as the target's state matrix at the current time; or, to determine the second association probability at the current time, where the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the target's state parameters; to multiply the second association probability at the current time with the predicted state matrix at the current time to obtain the third estimated state matrix at the current time; and to use the sum of the second estimated state matrix at the current time and the third estimated state matrix at the current time as the target's state matrix at the current time.

[0203] Optionally, the acquisition module 61 is used to determine several original observation state matrices at the current time based on the point cloud data at the current time; and to select the original observation state matrices associated with the target from the several original observation state matrices at the current time as several first observation state matrices at the current time.

[0204] Optionally, the association probability determination module 64 is further configured to determine the second association probability at the current time and the second association probability at the previous time, wherein the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target. The state update module 63 is configured to, in response to the second association probability at the current time being equal to a preset probability threshold, use the predicted state matrix at the current time as the state matrix of the target at the current time; in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being equal to the preset probability threshold, update the state of the target at the current time based on several first observation state matrices at the current time; in response to the second association probability at the current time being less than the preset probability threshold, and the second association probability at the previous time being less than the preset probability threshold, execute the determination of the velocity information of the target detection boxes corresponding to each first observation state matrix at the current time and subsequent steps thereof.

[0205] It should be noted that the apparatus of this embodiment can perform the steps in the above method. For a detailed description of the relevant content, please refer to the method section above, which will not be repeated here.

[0206] Please see Figure 7 , Figure 7 This is a schematic diagram of an embodiment of the electronic device provided in this application. In this embodiment, the electronic device 70 includes a memory 71 and a processor 72.

[0207] Processor 72 can also be referred to as CPU (Central Processing Unit). Processor 72 may be an integrated circuit chip with signal processing capabilities. Processor 72 can also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. A general-purpose processor can be a microprocessor, or processor 72 can be any conventional processor 72, etc.

[0208] The memory 71 in the electronic device 70 is used to store the program instructions required for the processor 72 to run.

[0209] The processor 72 is used to execute program instructions to implement the target tracking method in this application.

[0210] Please see Figure 8 , Figure 8This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium provided in this application. The computer-readable storage medium 80 of this embodiment stores program instructions 81, which, when executed, implement the target tracking method provided in this application. The program instructions 81 can form a program file and be stored in the aforementioned computer-readable storage medium 80 in the form of a software product, so that a computer device (which may be a personal computer, server, or network device, etc.) can execute all or part of the steps of the methods of various embodiments of this application. The aforementioned computer-readable storage medium 80 includes various media capable of storing program code, such as a USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or terminal devices such as computers, servers, mobile phones, and tablets.

[0211] The above scheme obtains several first observation state matrices based on point cloud data detected by a detection device at the current moment. It then determines the velocity information of the target detection boxes corresponding to each first observation state matrix at the current moment, and expands the dimension of the corresponding first observation state matrices based on the velocity information to obtain several second observation state matrices at the current moment. Finally, it updates the target's current state based on these second observation state matrices. The first observation state matrix includes the first state parameters of the corresponding target detection boxes, and the second observation state matrix includes the second state parameters of the corresponding target detection boxes. The second state parameters include the first state parameters and the velocity information. Since the detection device cannot directly detect the target's velocity information, the several first observation state matrices at the current moment lack velocity information, resulting in low accuracy for updating the target's current state directly based on these first observation state matrices. Therefore, updating the target's current state based on the velocity-expanded second observation state matrices at the current moment improves the accuracy of target tracking.

[0212] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0213] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0214] In the several embodiments provided in this application, it should be understood that the disclosed methods, apparatuses, and systems can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0215] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0216] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0217] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0218] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A target tracking method, characterized in that, The method includes: Based on the point cloud data detected by the detection device at the current moment, several first observation state matrices are obtained at the current moment, and the first observation state matrix includes the first state parameters of the corresponding target detection box. Determine at least one first observation state matrix from the previous time step; For each of the first observation state matrices at the current time, point cloud registration processing is performed on the point cloud data in the target detection box corresponding to the first observation state matrix at the current time and the point cloud data in the target detection box corresponding to the at least one first observation state matrix at the previous time to obtain at least one translation parameter. The derivative of the at least one translation parameter is calculated to obtain at least one first velocity. The velocity information of the target detection box corresponding to the first observation state matrix at the current time is obtained by combining the at least one first velocity. Based on the velocity information, the corresponding first observation state matrix at the current time is expanded to obtain several second observation state matrices at the current time. The second observation state matrix includes the second state parameters of the corresponding target detection box, and the second state parameters include the first state parameters and the velocity information. Based on the several second observation state matrices at the current moment, update the current state of the target.

2. The method according to claim 1, characterized in that, Before determining the velocity information of the target detection boxes corresponding to each of the first observation state matrices at the current time, the method further includes: Determine the first association probability corresponding to each of the first observation state matrices at the previous time step. The first association probability is used to represent the probability that the state parameter of the first observation state matrix is ​​the state parameter of the target. Determining at least one first observation state matrix from the previous time step includes: Based on the first association probability corresponding to each of the first observation state matrices at the previous time step, the at least one first observation state matrix at the previous time step is determined.

3. The method according to claim 2, characterized in that, Determining the at least one first observation state matrix at the previous time step based on the first association probabilities corresponding to each of the first observation state matrices at the previous time step includes: The first observation state matrix with the highest first correlation probability at the previous time step is used as the at least one first observation state matrix at the previous time step.

4. The method according to claim 1, characterized in that, The process of updating the target's current state based on the plurality of second observation state matrices at the current time includes: Kalman state updates are performed based on each of the second observation state matrices at the current time to obtain the first estimated state matrix corresponding to each of the second observation state matrices at the current time. Based on the first association probability corresponding to each of the second observation state matrices at the current time, the first estimated state matrices at the current time are weighted to obtain the second estimated state matrix at the current time. The second estimated state matrix at the current time is used as the state matrix of the target at the current time; or, a second association probability is determined at the current time, which represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target; the second association probability at the current time is multiplied by the predicted state matrix at the current time to obtain a third estimated state matrix at the current time; the sum of the second estimated state matrix at the current time and the third estimated state matrix at the current time is used as the state matrix of the target at the current time.

5. The method according to claim 1, characterized in that, The point cloud data obtained from the detection device at the current moment is used to acquire several first observation state matrices at the current moment, including: Based on the point cloud data at the current moment, determine several original observation state matrices at the current moment; From the plurality of original observation state matrices at the current moment, the original observation state matrix associated with the target is selected as the plurality of first observation state matrices at the current moment.

6. The method according to claim 5, characterized in that, The method further includes: Determine the second association probability at the current time and the second association probability at the previous time, where the second association probability represents the probability that the state parameters of the predicted state matrix obtained based on Kalman prediction are the state parameters of the target; In response to the second association probability at the current moment being equal to a preset probability threshold, the predicted state matrix at the current moment is used as the state matrix of the target at the current moment; In response to the fact that the second association probability at the current moment is less than the preset probability threshold, and the second association probability at the previous moment is equal to the preset probability threshold, the state of the target at the current moment is updated based on the plurality of first observation state matrices at the current moment. In response to the fact that the second association probability at the current moment is less than the preset probability threshold, and the second association probability at the previous moment is less than the preset probability threshold, the steps of determining the velocity information of the target detection boxes corresponding to each of the first observation state matrices at the current moment and subsequent steps are executed.

7. A target tracking device, characterized in that, The device includes: The acquisition module is used to acquire several first observation state matrices at the current moment based on the point cloud data detected by the detection device at the current moment. The first observation state matrix includes the first state parameters of the corresponding target detection box. A velocity expansion module is used to determine at least one first observation state matrix from the previous time step; for each first observation state matrix at the current time step, point cloud registration processing is performed on the point cloud data in the target detection box corresponding to the first observation state matrix at the current time step and the point cloud data in the target detection box corresponding to the at least one first observation state matrix from the previous time step, respectively, to obtain at least one translation parameter; the derivative of the at least one translation parameter is calculated to obtain at least one first velocity; and the velocity information of the target detection box corresponding to the first observation state matrix at the current time step is obtained by combining the at least one first velocity; the corresponding first observation state matrix at the current time step is expanded based on the velocity information to obtain several second observation state matrices at the current time step, wherein the second observation state matrix includes the second state parameter of the corresponding target detection box step, and the second state parameter includes the first state parameter and the velocity information; The state update module is used to update the current state of the target based on the plurality of second observation state matrices at the current time.

8. An electronic device, characterized in that, Including interconnected memory and processor, The memory stores program instructions; The processor is used to execute program instructions stored in the memory to implement the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program instructions that can be executed by a processor to implement the method of any one of claims 1-6.

Images