Multi-target tracking method and device of millimeter wave radar and storage medium

Abstract

This application discloses a multi-target tracking method, device, and storage medium for millimeter-wave radar, relating to the field of radar signal technology. The method includes: obtaining the determinant values ​​of the observation error covariance matrices of each sub-filter under its corresponding motion model, and comparing their ratios. A larger ratio indicates a higher confidence level in the predicted value of the sub-filter under the corresponding motion model, and the motion state description of the motion model is more consistent with the radar signal, avoiding the problem of computational model probability failure caused by independent predictions by two sub-filters; simultaneously, establishing a unified association gate for the motion model, extracting effective point cloud data as unified point cloud data for the motion model, and establishing an acknowledgment matrix for performing joint probability data association, thereby unifying the joint probability data association calculation of multiple sub-filters into a single operation, reducing computational resources.

Description

Technical Field

[0001] This application relates to the field of radar signal technology, and in particular to multi-target tracking methods, devices and storage media for millimeter-wave radar. Background Technology

[0002] Target tracking is a core technology of radar signal processing, playing a crucial role, especially in intelligent transportation systems. It can predict and update the vehicle's motion state and trajectory by analyzing target information detected by millimeter-wave radar, such as distance, speed, and azimuth, thereby enabling functions such as obstacle detection and avoidance, and adaptive cruise control.

[0003] Multi-target tracking technology can be broadly categorized into vehicle state estimation and data association. Vehicle state estimation typically employs the Interacting Multiple Mode (IMM) algorithm, which estimates the vehicle state using multiple models and then fuses the results in a specific ratio to obtain a more robust estimation model. Data association often utilizes the Joint Probability Data Association (JPDA) algorithm for tracking multiple targets in cluttered environments. Therefore, currently, IMM models and JPDA algorithms are commonly combined to achieve multi-target tracking.

[0004] However, in the traditional method of combining the IMM model and the JPDA algorithm, each sub-filter in the IMM filter is predicted independently, which leads to different echoes of each sub-filter, resulting in the failure of the likelihood function calculation model probability. In addition, each sub-filter needs to perform data correlation separately, which consumes too much computing resources. Summary of the Invention

[0005] The main purpose of this application is to provide a multi-target tracking method, device and storage medium for millimeter-wave radar, which aims to solve the technical problems of computational model probability failure and large computational resource requirements in the traditional method of combining the IMM model and JPDA algorithm.

[0006] To achieve the above objectives, this application proposes a multi-target tracking method for millimeter-wave radar. The millimeter-wave radar includes multiple target trackers, one of which tracks a radar signal. Each target tracker includes a first sub-filter and a second sub-filter. The multi-target tracking method for millimeter-wave radar includes:

[0007] The first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained. The ratio of the first determinant value and the second determinant value is compared to obtain the target determinant value with the larger ratio.

[0008] The motion model corresponding to the target determinant value is used as a reference model. A unified association gate for the uniform acceleration motion model and the uniform turning motion model is established based on the reference model. The motion model is either the uniform acceleration motion model or the uniform turning motion model.

[0009] Through the association gate, valid point cloud data and first point cloud data corresponding to the valid point cloud data are obtained, and an confirmation matrix is ​​established based on the association relationship between the valid point cloud data and the first point cloud data.

[0010] Based on the confirmation matrix, joint probability data association is performed to obtain the target motion state and target covariance matrix of each tracked target.

[0011] In one embodiment, before the step of obtaining the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model, the method includes:

[0012] Multiple sliding windows are set up to store multiple radar signals output by the millimeter-wave radar in the corresponding sliding windows. When a sliding window operation is detected to be completed, the multiple radar signals corresponding to the currently completed sliding window operation are extracted to obtain multiple frames of target tracking data.

[0013] Density-based clustering is performed on the multi-frame tracking target data to obtain a cloud of points to be matched.

[0014] In one embodiment, after the step of performing density-based clustering processing on the multi-frame tracking target data to obtain a cloud of points to be matched, the method includes:

[0015] Initialize each of the target trackers;

[0016] The motion states corresponding to the uniform acceleration motion model and the uniform turning motion model are respectively input into the initialized target tracker for interaction, so as to obtain the weighted total estimated state and weighted total estimated state covariance matrix of the uniform acceleration motion estimation state and the uniform turning motion estimation state of the first sub-filter, and the weighted total estimated state and weighted total estimated state covariance of the uniform acceleration motion estimation state and the uniform turning motion estimation state of the second sub-filter.

[0017] The target trajectory corresponding to each target point cloud data in the target point cloud set of the previous moment is traversed. Under the uniform acceleration motion model and the uniform turning motion model, the first sub-filter and the second sub-filter corresponding to each target trajectory are predicted and calculated respectively. The predicted motion state value and the predicted state covariance matrix value at the current moment are calculated to obtain the predicted point cloud set.

[0018] In one embodiment, the step of establishing a unified association gate for the uniform acceleration motion model and the uniform turning motion model based on the reference model includes:

[0019] The point cloud data to be matched is traversed, and the predicted point cloud data corresponding to the point cloud data to be matched is traversed according to the traversed point cloud data to be matched, so as to obtain the target predicted point cloud data corresponding to each point cloud data to be matched.

[0020] Based on the reference module, the association gate is established on each of the target prediction point cloud data.

[0021] In one embodiment, the step of obtaining valid point cloud data through the association gate and establishing an confirmation matrix based on the association relationship between the valid point cloud data and the first point cloud data includes:

[0022] The association gate is used to determine whether the target point cloud data falls within the association range of the association gate.

[0023] If the target point cloud data falls within the associated range, the point cloud data to be matched corresponding to the target point cloud data falling within the associated range will be extracted as the valid point cloud data.

[0024] An initial confirmation matrix is ​​established based on the correlation between the effective point cloud data and the first point cloud data;

[0025] Obtain multiple predicted point cloud indices associated with each valid point cloud data in the initial confirmation matrix, group valid point cloud data with the same predicted point cloud index into a group, and obtain multiple groups, wherein each group is a confirmation matrix.

[0026] In one embodiment, the step of performing joint probability data association based on the confirmation matrix to obtain the target motion state and target covariance matrix of each tracked target includes:

[0027] The confirmation matrix is ​​traversed and split to obtain multiple group confirmation matrices. Then, each group confirmation matrix is ​​split to obtain an interconnection matrix, where each interconnection matrix represents a feasible joint event.

[0028] The probability of the feasible joint event is calculated based on the interconnection matrix to obtain first probability data, and the probability that the valid point cloud data included in the interconnection matrix belongs to the first point cloud data is calculated to obtain second probability data.

[0029] Based on the comparison relationship between the second probability data and the preset probability threshold, if the second probability data is greater than the preset probability threshold, the matching status of the valid point cloud data corresponding to the second probability data is marked as a correct matching status. The first point cloud data in the correct matching status is determined as the first point cloud data to be calculated, and the valid point cloud data in the correct matching status is determined as the second point cloud data to be calculated.

[0030] Calculate the Kalman gain and innovation vector of the first point cloud data to be calculated and the second point cloud data to be calculated under the uniform acceleration motion model and the uniform turning motion model, respectively. Based on the Kalman gain, the innovation vector and the first point cloud data to be calculated, update the motion state prediction value and the state covariance matrix prediction value to obtain the next motion state update prediction value and the next state covariance matrix update prediction value.

[0031] The model matching probability at the current moment is calculated using the maximum likelihood function;

[0032] Based on the model matching probability, the predicted value of the next motion state update and the predicted value of the next state covariance matrix update are fused to obtain the target weighted motion state and the target weighted covariance matrix.

[0033] In one embodiment, after the step of marking the matching status of the valid point cloud data corresponding to the second probability data as a correct matching status if the second probability data is greater than the second preset probability threshold, the method includes:

[0034] Valid point cloud data other than the second point cloud data to be calculated are identified as point cloud data to be judged with an incorrect matching status;

[0035] Determine whether the point cloud data to be judged meets the preset creation conditions;

[0036] If the point cloud data to be judged meets the preset creation conditions, then the point cloud data to be judged is determined as the first next target point cloud data in the next target tracking process.

[0037] In one embodiment, after the step of obtaining the target motion state and target covariance matrix of each tracked target, the method includes:

[0038] The track and point cloud attributes corresponding to each first point cloud data are traversed. First point cloud data with a track status of dead track are extracted from each track, and first point cloud data with an attribute status of abnormal attribute are extracted from each point cloud attribute. Then, the first point cloud data with a track status of dead track and / or the first point cloud data with an attribute status of abnormal attribute are removed to obtain the second next target point cloud data.

[0039] The first next target point cloud data and the second next target point cloud data are input into the next target tracking process to perform calculations for each of the tracked targets.

[0040] In addition, to achieve the above objectives, this application also proposes a multi-target tracking device for millimeter-wave radar, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multi-target tracking method for millimeter-wave radar as described above.

[0041] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the multi-target tracking method of millimeter-wave radar as described above.

[0042] One or more technical solutions proposed in this application have at least the following technical effects:

[0043] The first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained. The ratio of the first and second determinant values ​​is compared to obtain the target determinant value with the larger ratio. The motion model corresponding to the target determinant value is used as a reference model. A unified correlation gate for the uniform acceleration motion model and the uniform turning motion model is established based on the reference model, where the motion model is either a uniform acceleration motion model or a uniform turning motion model. Through the correlation gate, effective point cloud data is extracted from the point cloud set to be matched, and first point cloud data corresponding to the effective point cloud data is extracted from the predicted point cloud set. A confirmation matrix is ​​established based on the correlation between the effective point cloud data and the first point cloud data. Joint probability data correlation is performed based on the confirmation matrix to obtain the target motion state and target covariance matrix of each tracked target.

[0044] This application establishes a unified correlation gate for uniform acceleration motion and uniform turning motion models by using the motion model corresponding to the target determinant value as a reference model. This is because both the first and second sub-filters are used to process radar signals. The difference is that the first sub-filter describes the uniform acceleration motion state of the radar signal under the uniform acceleration motion model, while the second sub-filter describes the uniform turning motion state of the radar signal under the uniform turning motion model. Therefore, the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained. The first and second determinant values ​​are then compared. The determinant values ​​are compared by ratio. A larger ratio indicates that the sub-filter has a higher confidence level in the predicted value under the motion model corresponding to that determinant value. The motion state description of the motion model is more consistent with the radar signal, avoiding the problem of probability calculation model failure caused by the two sub-filters making independent predictions. At the same time, a unified association gate is established for the uniform acceleration motion model and the uniform turning motion model based on the reference model. The effective point cloud data is extracted from the point cloud set to be matched through the association gate and applied to the unified point cloud data of the uniform acceleration motion model and the uniform turning motion model. An acknowledgment matrix is ​​established to perform joint probability data association. In this way, the joint probability data association calculation of multiple sub-filters is unified into one, reducing computational resources. Attached Figure Description

[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating an embodiment of the multi-target tracking method for millimeter-wave radar provided in this application.

[0048] Figure 2 This is a schematic diagram of the overall process of the multi-target tracking method for millimeter-wave radar in this application;

[0049] Figure 3 This is a flowchart illustrating Embodiment 2 of the multi-target tracking method for millimeter-wave radar in this application;

[0050] Figure 4 This is a flowchart illustrating Embodiment 3 of the multi-target tracking method for millimeter-wave radar provided in this application;

[0051] Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the multi-target tracking method of millimeter-wave radar in the embodiments of this application.

[0052] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0053] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0054] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0055] The main solution of this application embodiment is as follows: First, obtain the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model, and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model. Compare the ratio of the first and second determinant values ​​to obtain the target determinant value with the larger ratio. Use the motion model corresponding to the target determinant value as a reference model. Establish a unified association gate for the uniform acceleration motion model and the uniform turning motion model based on the reference model, where the motion model is either a uniform acceleration motion model or a uniform turning motion model. Through the association gate, obtain effective point cloud data and the first point cloud data corresponding to the effective point cloud data. Establish a confirmation matrix based on the association relationship between the effective point cloud data and the first point cloud data. Perform joint probability data association based on the confirmation matrix to obtain the target motion state and target covariance matrix of each tracked target.

[0056] Because in the existing traditional method of combining the IMM model and the JPDA algorithm, each sub-filter in the IMM filter is predicted independently, resulting in different echoes from each sub-filter. This leads to the failure of the likelihood function calculation model probability. In addition, each sub-filter needs to perform data correlation independently, which consumes too much computational resources.

[0057] This application provides a solution that establishes a unified correlation gate for uniform acceleration motion model and uniform turning motion model by using the motion model corresponding to the target determinant value as a reference model. This is because both the first and second sub-filters are used to process radar signals. The difference is that the first sub-filter describes the uniform acceleration motion state of the radar signal under the uniform acceleration motion model, while the second sub-filter describes the uniform turning motion state of the radar signal under the uniform turning motion model. Therefore, the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained, and the first determinant value and... The second determinant value is compared by ratio. A larger determinant value indicates that the sub-filter has a higher confidence level in the predicted value under the motion model corresponding to that determinant value. The motion state description of the motion model is more consistent with the radar signal, avoiding the problem of probability calculation model failure caused by the two sub-filters making independent predictions. At the same time, a unified association gate for uniform acceleration motion model and uniform turning motion model is established based on the reference model. The effective point cloud data is extracted from the point cloud set to be matched through the association gate as the unified point cloud data for uniform acceleration motion model and uniform turning motion model. An acknowledgment matrix is ​​established to perform joint probability data association. In this way, the joint probability data association calculation of multiple sub-filters is unified into one, reducing computational resources.

[0058] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as a millimeter-wave radar multi-target tracking device. The following description uses a millimeter-wave radar multi-target tracking device as an example to illustrate this embodiment and the subsequent embodiments.

[0059] Based on this, embodiments of this application provide a multi-target tracking method for millimeter-wave radar, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the multi-target tracking method for millimeter-wave radar of this application.

[0060] In this embodiment, the multi-target tracking method of the millimeter-wave radar includes steps S10 to S40:

[0061] Step S10: Obtain the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model, and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model, and compare the ratio of the first determinant value and the second determinant value to obtain the target determinant value with the larger ratio.

[0062] It should be noted that millimeter-wave radar includes multiple target trackers. One target tracker is used to track one radar signal (i.e., point cloud data). One target tracker includes a first sub-filter and a second sub-filter. The following description uses one target tracker.

[0063] In conventional methods that combine the IMM model and the JPDA algorithm, probability data is calculated based on the echoes of radar signals under various operating models obtained from the sub-filters in the target tracker. However, since a radar signal corresponds to only one motion state, i.e., a radar signal corresponds to only a uniform acceleration motion state or a uniform turning motion state, estimating a radar signal by using the motion model corresponding to each sub-filter results in low accuracy in describing the motion state of the radar signal. The two sub-filters will obtain different echoes, and different echoes will cause the probability calculation model to fail.

[0064] To avoid the aforementioned situation, this embodiment proposes to acquire only the motion model that more accurately describes the motion state of the radar signal. Specifically, firstly, the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model is calculated, and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model is calculated. Then, the ratio of the first determinant value and the second determinant value is compared, and the target determinant value with the larger ratio is obtained. Assuming that the first determinant value is greater than the second determinant value during the ratio comparison, the first determinant value is determined as the target determinant value.

[0065] The reason for comparing the ratios of the determinant values ​​is that a larger determinant value indicates a higher degree of confidence in the predicted value of the motion model corresponding to that determinant value, meaning that the motion model more accurately describes the motion of the radar signal. For example, if the first determinant value is the target determinant value, it means that the confidence in the predicted value of the observation error covariance matrix of the uniform acceleration motion model corresponding to the first determinant value is higher. In this case, the radar signal is undergoing uniform acceleration, so the uniform acceleration motion model more accurately describes the motion state of the radar signal, thus avoiding interference from the uniform turning motion model.

[0066] Step S20: Take the motion model corresponding to the target determinant value as a reference model, and establish a unified association gate for the uniform acceleration motion model and the uniform turning motion model based on the reference model, wherein the motion model is the uniform acceleration motion model or the uniform turning motion model.

[0067] After determining the motion model through step S10, the motion model is used as a reference model to establish a unified association gate for the uniform acceleration motion model and the uniform turning motion model, thereby unifying the point cloud data of the uniform acceleration motion model and the uniform turning motion model.

[0068] In one feasible implementation, step S20 includes steps S21-S22:

[0069] Step S21: Traverse each of the point cloud data to be matched, and based on the traversed point cloud data to be matched, traverse the predicted point cloud data corresponding to the point cloud data to be matched, to obtain the target predicted point cloud data corresponding to each point cloud data to be matched.

[0070] Step S22, based on the reference module, establish the association gate (i.e., ...) on each of the target prediction point cloud data. Figure 2 Step S8 in the process.

[0071] Since not every radar signal undergoes uniform acceleration or uniform turning motion, processing each radar signal in step S10 yields multiple uniform acceleration and uniform turning motion models. These models are then used as reference models to establish a unified association gate. Specifically, by traversing each point cloud data set to be matched, and during this traversal, traversing the predicted point cloud data corresponding to each point cloud data to be matched, the unified process is as shown in expressions ① to ④.

[0072] ———— Expression ①

[0073] ————Expression ②

[0074] ————Expression ③

[0075] ————Expression ④

[0076] Expression ① represents the observation residual covariance matrix of the m-th motion model for the t-th predicted point cloud data, where H represents the observation matrix. Expression ② represents the observed predicted value of the t-th predicted point cloud data at time K-1 (i.e., the previous time) after being updated with the m-th motion model information, for time K (i.e., the current time). Expression ③ represents the prediction information vector between the j-th point cloud data to be matched and the t-th predicted point cloud data. Expression ④ represents the correlation gate, used to determine whether the point cloud data to be matched is valid. It should be noted that in this embodiment, m=2, meaning it includes both uniform acceleration motion models and uniform turning motion models.

[0077] The observation error covariance matrix output by the first sub-filter under the uniformly accelerated motion model is input into expression ①, which outputs the observation residual covariance matrix of the predicted point cloud data under the uniformly accelerated motion model. Simultaneously, based on the predicted point cloud data at time K-1, the observation prediction value at time K is updated using the uniformly accelerated motion model. Then, based on this observation prediction value, the prediction innovation vector of the predicted point cloud data and the corresponding target point cloud data is obtained through expression ③. Substituting this prediction innovation vector and the corresponding observation residual covariance matrix into expression ④ yields the judgment value of the target point cloud data. Determine whether the judgment value is within the valid range. If the judgment value is within the valid range, it means that the motion state of the point cloud data to be matched is consistent with the motion state of the current multi-target tracking.

[0078] Step S30: Obtain valid point cloud data and first point cloud data corresponding to the valid point cloud data through the association gate, and establish a confirmation matrix based on the association relationship between the valid point cloud data and the first point cloud data.

[0079] Through the association gate in step S20 above, if the judgment value corresponding to the target point cloud data is within the valid range (i.e. Figure 2 In step S9), the point cloud data to be matched corresponding to the target point cloud data is determined as valid point cloud data. Since the predicted point cloud data is the predicted data of the point cloud data to be matched at a new time, the predicted point cloud data at the new time has a corresponding relationship with the point cloud data to be matched. At this time, the predicted point cloud data corresponding to the valid point cloud data is obtained, and after the predicted point cloud data is determined as the first point cloud data, a confirmation matrix is ​​established based on the correlation between the valid point cloud data and the first point cloud data. Based on the confirmation matrix, the complex calculation process is simplified, saving calculation time and effort and improving calculation efficiency.

[0080] In one feasible implementation, step S30 may include steps S31-S32:

[0081] Step S31: Determine whether the target point cloud data falls within the association range of the association gate through the association gate.

[0082] Step S32: If the target point cloud data falls within the associated range, the point cloud data to be matched corresponding to the target point cloud data falling within the associated range is extracted as the valid point cloud data.

[0083] As can be seen from expressions ① to ④, the association gate is used to determine whether the target point cloud data is within the association range of the association gate established with the point cloud data to be matched associated with it, that is, whether the judgment value corresponding to the target point cloud data is within the valid range. If the target point cloud data is within the association range, the point cloud data to be matched associated with it is determined as valid point cloud data.

[0084] If the target point cloud data is not within the associated range, it means that the vehicle motion state corresponding to the target point cloud data does not conform to the current multi-target tracking motion state and there is an abnormal motion state. Therefore, the point cloud data to be matched associated with it is determined as abnormal point cloud data and will not be included in the execution calculation of subsequent process steps.

[0085] Step S33: Establish an initial confirmation matrix based on the correlation between the effective point cloud data and the first point cloud data.

[0086] Step S34: Obtain multiple predicted point cloud indices associated with each valid point cloud data in the initial confirmation matrix, group the valid point cloud data with the same predicted point cloud index into one group, and obtain multiple groups, wherein one group is one confirmation matrix.

[0087] Based on the correlation between the valid point cloud data and the first point cloud data, an initial confirmation matrix (i.e., Figure 2 After step S10), the initial confirmation matrix is ​​grouped to obtain the confirmation matrix (i.e., Figure 2 Step S11 in the process is as follows:

[0088] First, based on each valid point cloud data in each initial confirmation matrix, multiple predicted point cloud indices associated with each valid point cloud data are confirmed. Then, each valid point cloud data is traversed. If two valid point cloud data are found to have a common predicted point cloud index, the two valid point cloud data are grouped together. The set of predicted point cloud indices of this group is the union of the common predicted point cloud indices between the two valid point cloud data. Then, other valid point cloud data are traversed to confirm whether there are any valid point cloud data belonging to the same predicted point cloud index as the set of predicted point cloud indices of this group. If so, the valid point cloud data is assigned to this group. This process is repeated until all valid point cloud data are divided into multiple groups. Confirmation matrices are established for each of the multiple groups. The valid point cloud data in each confirmation matrix are associated with other valid point cloud data in the confirmation matrix through their associated predicted point cloud indices.

[0089] Explain using expression ⑤ as an example.

[0090] ————Expression ⑤

[0091] in, Let represent the confirmation matrix in row j and column t, where n indicates which group the confirmation matrix belongs to. Expression ⑤ determines whether the valid point cloud data in the confirmation matrix is ​​correlated with the first point cloud data. Specifically:

[0092] During the traversal of each valid point cloud data, it is determined whether the predicted point cloud data of the t-th first point cloud data is associated with the j-th valid point cloud data. If so, expression ⑤ outputs 1; otherwise, expression ⑤ outputs 0. Grouping involves splitting the target confirmation matrix into multiple smaller confirmation matrices.

[0093] The final multiple confirmation matrices only establish the correlation between the first predicted point cloud data within the group and the predicted point cloud data within the group. The first column in the confirmation matrix indicates that the valid point cloud data may originate from clutter.

[0094] Step S40: Perform joint probability data association based on the confirmation matrix to obtain the target motion state and target covariance matrix of each tracked target.

[0095] The confirmation matrix is ​​used to perform joint probability data association (i.e. Figure 2 In step S13), the target motion state and target covariance matrix of each tracked target are obtained, thereby unifying the joint probability data of multiple sub-filters into a single set, reducing computational resources. The details are as follows:

[0096] Step S40 includes steps S41 to S46:

[0097] Step S41: Traverse each of the confirmation matrices and split each of the confirmation matrices to obtain multiple group confirmation matrices. Then, split each of the group confirmation matrices to obtain interconnection matrices, wherein each interconnection matrix represents a feasible joint event.

[0098] The confirmation matrix is ​​traversed, and each confirmation matrix is ​​then split to obtain multiple grouped confirmation matrices (i.e., Figure 2 After step S12), the confirmation matrix of each group is split to obtain several interconnection matrices, each interconnection matrix representing a feasible connection event. Where L is the sum of feasible joint events. Figure 2 One of the Groups is a group confirmation matrix.

[0099] Step S42: Calculate the probability of the feasible joint event based on the interconnection matrix to obtain first probability data, and calculate the probability that the valid point cloud data included in the interconnection matrix belongs to the first point cloud data to obtain second probability data.

[0100] Calculate the probability of feasible joint events based on expressions ① and ⑥~⑦.

[0101] ———— Expression ①

[0102] ————Expression ⑥

[0103] —expression⑦

[0104] Where T represents the number of valid point cloud data in the feasible joint events. This represents the valid point cloud data at time K. This represents the probability of a correct detection within the valid point cloud. This indicates the number of valid point cloud data items in the executable joint event that did not match the predicted point cloud index. This represents the information vector of the valid point cloud data and its associated first point cloud data. Let M represent the observation error covariance matrix, and M represent the dimension of the innovation vector. This represents the first probability data.

[0105] After calculating the first probability data based on expressions ① and ⑥~⑦, the second probability data is calculated based on expression ⑧.

[0106] ————Expression ⑧

[0107] in, This indicates that the j-th valid point cloud data belongs to the second probability data of the first point cloud data.

[0108] Step S43: Based on the comparison relationship between the second probability data and the preset probability threshold, if the second probability data is greater than the preset probability threshold, then the matching status of the valid point cloud data corresponding to the second probability data is marked as a correct matching status. In this case, the first point cloud data in the correct matching status is determined as the first point cloud data to be calculated, and the valid point cloud data in the correct matching status is determined as the second point cloud data to be calculated.

[0109] Step S44: Calculate the Kalman gain and innovation vector of the first point cloud data to be calculated and the second point cloud data to be calculated under the uniform acceleration motion model and the uniform turning motion model, respectively. Based on the Kalman gain, the innovation vector and the first point cloud data to be calculated, update the motion state prediction value and the state covariance matrix prediction value to obtain the next motion state update prediction value and the next state covariance matrix update prediction value.

[0110] Based on the calculated second probability data, if it is determined that the second probability data is greater than the second preset probability threshold, then the matching state of the valid point cloud data corresponding to the second probability data is determined to be the correct matching state. After the valid point cloud data with the correct matching state is determined as the second point cloud data to be calculated, the Kalman gain and innovation vector of the first point cloud data to be calculated and the second point cloud data to be calculated under the uniform acceleration motion model and the uniform turning motion model, respectively, are calculated based on the second probability data and expressions ⑨ and ⑩.

[0111] ———— Expression 9

[0112] ———— Expression ⑩

[0113] Where n represents the number of valid point clouds in the confirmation matrix, This represents the innovation vector of the t-th point cloud data under the m-th motion model after calculating and integrating the covariance matrices of all observation errors. Let represent the Kalman gain of the t-th point cloud data under the m-th motion model.

[0114] At this point, we iterate through each first point cloud data and valid point cloud data. If the matching status of the iterated first point cloud data / valid point cloud data is a correct match, that is, if we obtain the first point cloud data to be calculated and the second point cloud data to be calculated, then we base the expression on the first point cloud data and the second point cloud data to be calculated. ~ And, in conjunction with the Kalman gain and the innovation vector, update the predicted motion state values ​​and the predicted state covariance matrix values ​​obtained in step A50 (i.e., Figure 2 Step S14 in the process.

[0115] ----expression

[0116]

[0117] ---expression

[0118] in, Update the predicted value for the next motion state after the update at time K. The updated prediction value is the next state covariance matrix after the update at time K.

[0119] Step S45: Calculate the model matching probability at the current time using the maximum likelihood function (i.e., Figure 2 Step S15 in the process.

[0120] After the uniform acceleration motion model and the uniform turning motion model estimate the predicted values ​​of the motion state and the state covariance matrix, respectively, the maximum likelihood function (i.e., the expression) is used. Calculate the similarity of the current valid point cloud data to obtain the weight of the fit between the uniform acceleration motion model and the uniform turning motion model and the target tracker, i.e., the likelihood value.

[0121] ----expression

[0122] in, Let represent the likelihood value of the m-th motion model relative to the target tracker at time K (i.e., the current time).

[0123] Based on the likelihood values ​​of the uniform acceleration motion model and the uniform turning motion model obtained above, based on the expression Calculate the matching probability between each motion model and the target tracker at time K.

[0124] ----expression

[0125] in, Let represent the model matching probability of the m-th motion model at time K. and By expression respectively and expression The result.

[0126] ----expression

[0127] ----expression

[0128] Step S46: Based on the model matching probability, fuse the next motion state update prediction value and the next state covariance matrix update prediction value to obtain the target weighted motion state and the target weighted covariance matrix (i.e., Figure 2 Step S16 in the process.

[0129] Based on the model matching probability obtained in step S45, substitute the next motion state update prediction value or the next state covariance matrix update prediction value corresponding to the uniform acceleration motion model and the uniform turning motion model into the expression. or expression Specifically:

[0130] ----expression

[0131] --expression

[0132] Substitute the predicted values ​​of the next motion state update from the uniform acceleration motion model and the uniform turning motion model into the expression. The process involves fusing the data and substituting the next-state covariance matrix update predictions from the uniform acceleration motion model and the uniform turning motion model into the expression. By fusing the data, the target motion state can be obtained. and target covariance matrix .

[0133] and These are the optimal estimates of the target motion state and target covariance after weighting the two motion models, respectively.

[0134] In this embodiment, the motion model corresponding to the target determinant value is used as a reference model to establish a unified correlation gate for the uniform acceleration motion model and the uniform turning motion model. This is because both the first sub-filter and the second sub-filter are used to process radar signals. The difference is that the first sub-filter describes the uniform acceleration motion state of the radar signal under the uniform acceleration motion model, while the second sub-filter describes the uniform turning motion state of the radar signal under the uniform turning motion model. Therefore, the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained, and the first determinant value and the second determinant value are then compared. The determinant values ​​are compared by ratio. A larger determinant value indicates a higher confidence level in the predicted value of the sub-filter under the corresponding motion model. The motion state description of the motion model is more consistent with the radar signal, avoiding the problem of probability calculation model failure caused by independent prediction by two sub-filters. At the same time, a unified association gate is established for the uniform acceleration motion model and uniform turning motion model based on the reference model. The effective point cloud data is extracted from the point cloud set to be matched through the association gate as the unified point cloud data for the uniform acceleration motion model and uniform turning motion model. An acknowledgment matrix is ​​established to perform joint probability data association. In this way, the joint probability data association calculation of multiple sub-filters is unified into one, reducing computational resources.

[0135] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Combination Figure 2 Before step S10, the multi-target tracking method of the millimeter-wave radar further includes steps A10 to A20:

[0136] Step A10: Set multiple sliding windows, store multiple radar signals output by the millimeter-wave radar in the corresponding sliding windows, and when a sliding window operation is detected to be completed, extract the multiple radar signals corresponding to the currently completed sliding window operation to obtain multiple frames of target tracking data.

[0137] To reduce random fluctuations in radar signals and thus more clearly reveal the true trend of radar signals, this embodiment sets multiple sliding windows (i.e., Figure 2 Step S1) is used to store radar signals (i.e. Figure 2 (F1, F2, ..., Fn-2, Fn-1, Fn, Fn+1, Fn+2, ...). When a radar signal from a millimeter-wave radar is received at the current moment, the radar information is stored at the front of the sliding window. When the next moment arrives and a new radar signal from the millimeter-wave radar is received at the next moment, the position of the radar signal stored in the sliding window changes. The storage position of each frame of radar signal moves backward, and the new radar signal is stored in the storage position of the previous frame of radar signal, and so on, continuously sliding and storing. When the accumulated radar signals reach n frames, the radar signal in the last frame exits the sliding window, and the latest radar signal is stored in the frontmost sliding window. The radar signals located in the later sliding window storage positions need to perform position compensation based on the time interval between the new frame of radar signal and their own motion state each time a new frame of radar signal is stored. After the position compensation is completed, it is determined that one sliding window operation is completed.

[0138] Before extracting radar signals, to ensure the effectiveness of the extracted radar signals, each radar signal in the sliding window is filtered for outliers based on attributes such as SNRS, velocity, range, and RCS. Then, multiple adjacent radar signals are accumulated and extracted to obtain multi-frame tracking target data (i.e.,...). Figure 2 Step S2 in the process.

[0139] Step A20: Perform density-based clustering on the multi-frame tracking target data to obtain a cloud of points to be matched.

[0140] The point cloud data of each extracted multi-frame tracking target data is traversed, and the traversed point cloud data is marked as visited. Then, the neighborhood point cloud set of each point cloud data is searched (i.e., Figure 2Step S3). If the number of neighboring points is less than the minimum number of neighboring points, the point cloud data is marked as noise. If the number of neighboring points is greater than the minimum number of neighboring points, a new cluster is created, and the point cloud data and its neighboring point clouds are added to this cluster. Then, the cluster is expanded. For each point cloud in the neighboring point cloud set, if the point cloud data has not been visited, the point cloud data is traversed and marked as visited, and then the neighboring point cloud set of the point cloud data is searched. If the number of neighboring points is greater than or equal to the minimum number of neighboring points, the neighboring point set of the point cloud data is added to the neighboring point set of the cluster. If the number of neighboring points is greater than the minimum number of neighboring points, another new cluster is created, and so on, until all the point cloud data has been traversed and clustered to obtain multiple point cloud clusters. Each point cloud cluster is used as a set of point clouds to be matched (i.e., Figure 2 In step S4), the attribute information of the point cloud set to be matched is calculated by integrating the attribute information of all point cloud data within the point cloud cluster.

[0141] In one feasible implementation, after step A20, steps A30 to A50 are included:

[0142] Step A30: Initialize each of the target trackers.

[0143] If it is determined that the target tracker has not been initialized, then initialize the state transition matrix of the target tracker. ), the covariance matrix (R) of the observation noise, and the covariance matrix (Q) of the process noise of the system state transition model.

[0144] Step A40: Input the motion states corresponding to the uniform acceleration motion model and the uniform turning motion model into the initialized target tracker for interaction (i.e., Figure 2 In step S5), the weighted total estimated state and weighted total estimated state covariance matrix of the uniform acceleration motion estimation state and uniform turning motion estimation state of the first sub-filter are obtained, as well as the weighted total estimated state and weighted total estimated state covariance of the uniform acceleration motion estimation state and uniform turning motion estimation state of the second sub-filter.

[0145] ----expression

[0146] --expression

[0147] in, m=1,2, The scalar represents the correlation coefficient of the fusion, and the m-th row represents the correlation between all target trackers and the m-th target tracker; This represents the motion state of the m-th motion model at time K-1. This represents the transition matrix from target tracker m to target tracker j. This represents the model matching probability of sub-filter m at time K-1.

[0148] Specifically, based on expressions Calculate the weighted overall estimate of the uniform acceleration motion estimate state and the uniform turning motion estimate state of the first sub-filter. and based on expressions Calculate the weighted total estimated state covariance of the uniform acceleration motion estimation state and the uniform turning motion estimation state of the second sub-filter. .

[0149] Step A50: Iterate through the target trajectory corresponding to each target point cloud data in the target point cloud set of the previous moment. Under the uniform acceleration motion model and the uniform turning motion model respectively, perform prediction calculations on the first sub-filter and the second sub-filter corresponding to each target trajectory to obtain the predicted motion state value and the predicted state covariance matrix value at the current moment, thus obtaining the predicted point cloud set (i.e., Figure 2 Step S7 in the process.

[0150] One target point cloud data corresponds to one target trajectory of a vehicle. Therefore, the target trajectories corresponding to each target point cloud data in the acquired target point cloud set are traversed. During the traversal, based on the uniform acceleration motion model and the uniform turning motion model, the traversed target trajectories are input into their corresponding first and second sub-filters. The target trajectories input to the first sub-filter are processed by the uniform acceleration motion model, and the target trajectories input to the second sub-filter are processed by the uniform turning motion model. , , , , and Predicting six motion states (i.e.) Figure 2 Step S6 in the process, specifically, based on the expression .

[0151] ----expression

[0152] Furthermore, the state covariance matrix of the target trajectory input to the first sub-filter is predicted using a uniform acceleration motion model, and the state covariance matrix of the target trajectory input to the second sub-filter is predicted using a uniform turning motion model. Figure 2 Step S6 in the process, specifically, based on the expression .

[0153] ----expression

[0154] Where K represents time, This represents the predicted motion state at time K after the information update at time K-1 (i.e., the predicted motion state at the current time). This represents the predicted state covariance matrix at time K after the information update at time K-1 (i.e., the predicted state covariance matrix at the current time), where t is the target trajectory and m is the number of filters. In this example, the number of filters is the first sub-filter and the second sub-filter.

[0155] In this embodiment, a sliding window is set to reduce the random fluctuations of the radar signal, thereby more clearly showing the true trend of the point cloud data corresponding to the radar signal. At the same time, density-based clustering is used to extract the point cloud set to be matched, so as to obtain point cloud data with similar characteristics, so as to process the point cloud data with similar characteristics in subsequent steps.

[0156] Based on the first embodiment of this application, in the third embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 After step S43, the multi-target tracking method of the millimeter-wave radar further includes steps B11 to B13:

[0157] Step B11: Determine the valid point cloud data other than the second point cloud data to be calculated as point cloud data to be judged whose matching status is incorrect.

[0158] Step B12: Determine whether the point cloud data to be determined meets the preset creation conditions.

[0159] Step B13: If the point cloud data to be judged meets the preset creation conditions, then the point cloud data to be judged is determined as the first next target point cloud data in the next target tracking process.

[0160] It should be noted that steps S10 to S40 are continuously executed in a loop based on the multi-frame tracking targets output by the sliding window operation. The multi-target tracking operation performed based on the acquired multi-frame tracking targets is an operation to track the vehicle motion generated at the current moment. Because the correlation probability between the effective point cloud data and the first point cloud data at different moments is not the same, the effective point cloud data with a correct matching state is only the one at the current moment. The point cloud data to be judged whose matching state is determined to be incorrect needs to be judged whether it meets the target tracking conditions of the multi-target tracking operation at the next moment, that is, the preset creation conditions. The point cloud attributes of the point cloud data to be judged, such as whether the speed meets the speed requirements of the preset creation conditions corresponding to the next moment, are judged. If they meet the requirements, the point cloud data to be judged is input into the multi-target tracking operation at the next moment, that is, the point cloud data to be judged is determined as the first target point cloud data of the next target tracking process (i.e., Figure 2 Step S17 in the process is used to avoid the decrease in the consistency of multi-target tracking caused by the lack of target point cloud data.

[0161] Following step S40, the multi-target tracking method for millimeter-wave radar further includes steps S51-S52:

[0162] Step S51: Traverse the track and point cloud attributes corresponding to each first point cloud data, extract the first point cloud data with the track status of dead track from each track, and extract the first point cloud data with the attribute status of abnormal attribute from each point cloud attribute, and then remove the first point cloud data with the track status of dead track and / or the first point cloud data with the attribute status of abnormal attribute to obtain the second next target point cloud data.

[0163] Step S52: Input the first next target point cloud data and the second next target point cloud data into the next target tracking process to perform calculations for each of the tracked targets.

[0164] In this embodiment, by traversing the tracks and point cloud attributes corresponding to each first point cloud data, the first point cloud data that cannot be input into the multi-target tracking operation at the next moment is extracted. Specifically, each first point cloud data is traversed to determine whether its corresponding track overlaps with the tracks of other first point cloud data. If there is overlap, the track status of the track is determined to be a dead track. It is also determined whether the track is a track split from other vehicles. If so, the track status of the track is determined to be a dead track. Based on the point cloud attributes, it is determined whether the corresponding first point cloud data has track anomalies, such as speeding. If so, the attribute status of the point cloud data is determined to be abnormal. The first point cloud data with track status of dead track and / or attribute status of abnormal attribute are removed, and the remaining first point cloud data are used as the second next target point cloud data (i.e., Figure 2 Step S18), and the first next target point cloud data are input into the multi-target tracking operation at the next time step (i.e. Figure 2 Step S19 in the process.

[0165] In this embodiment, by filtering the valid point cloud data and the first point cloud data at the current moment, the first next target point cloud data and the second next target point cloud data that can be input into the multi-target tracking operation at the next moment are obtained, thereby ensuring the continuity of multi-target tracking.

[0166] This application provides a multi-target tracking device for millimeter-wave radar, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the multi-target tracking method for millimeter-wave radar in the first embodiment described above.

[0167] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of a multi-target tracking device suitable for implementing the embodiments of this application using millimeter-wave radar. The multi-target tracking device for millimeter-wave radar in the embodiments of this application may include, but is not limited to, mobile terminals such as laptops, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The multi-target tracking device for millimeter-wave radar shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0168] like Figure 5 As shown, the multi-target tracking device of millimeter-wave radar may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the millimeter-wave radar multi-target tracking device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the millimeter-wave radar multi-target tracking device to wirelessly or wiredly communicate with other devices to exchange data. Although a millimeter-wave radar multi-target tracking device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0169] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0170] The multi-target tracking device for millimeter-wave radar provided in this application employs the multi-target tracking method for millimeter-wave radar described in the above embodiments. This method addresses the technical problems of computational model probability failure and high computational resource requirements inherent in traditional methods combining the IMM model and the JPDA algorithm. Compared to existing technologies, the beneficial effects of the multi-target tracking device for millimeter-wave radar provided in this application are the same as those of the multi-target tracking method for millimeter-wave radar described in the above embodiments. Furthermore, other technical features of this multi-target tracking device for millimeter-wave radar are the same as those disclosed in the previous embodiment method, and will not be elaborated upon here.

[0171] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0172] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0173] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the multi-target tracking method of millimeter-wave radar in the above embodiments.

[0174] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0175] The aforementioned computer-readable storage medium may be included in a multi-target tracking device for millimeter-wave radar; or it may exist independently and not be assembled into a multi-target tracking device for millimeter-wave radar.

[0176] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a multi-target tracking device of a millimeter-wave radar, the multi-target tracking device of the millimeter-wave radar performs the following actions: It acquires the first determinant value of the first observation error covariance matrix of the first sub-filter under a uniform acceleration motion model, and the second determinant value of the second observation error covariance matrix of the second sub-filter under a uniform turning motion model. It then compares the ratio of the first and second determinant values ​​to obtain the target determinant value with the larger ratio. Using the motion model corresponding to the target determinant value as a reference model, it establishes a unified correlation gate for the uniform acceleration motion model and the uniform turning motion model based on the reference model. The motion model is either a uniform acceleration motion model or a uniform turning motion model. Through the correlation gate, it extracts effective point cloud data from the point cloud set to be matched, and extracts first point cloud data corresponding to the effective point cloud data from the predicted point cloud set. It establishes a confirmation matrix based on the correlation between the effective point cloud data and the first point cloud data. Based on the confirmation matrix, it performs joint probability data correlation to obtain the target motion state and target covariance matrix of each tracked target.

[0177] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0178] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0179] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0180] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the multi-target tracking method of the millimeter-wave radar described above. This solves the technical problems of computational model probability failure and large computational resource requirements inherent in traditional methods that combine the IMM model and the JPDA algorithm. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the multi-target tracking method of millimeter-wave radar provided in the above embodiments, and will not be elaborated upon here.

[0181] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A multi-target tracking method for millimeter-wave radar, characterized in that, The millimeter-wave radar includes multiple target trackers, one target tracker is used to track one radar signal, and each target tracker includes a first sub-filter and a second sub-filter. The multi-target tracking method of the millimeter-wave radar includes: The first determinant value of the first observation error covariance matrix of the first sub-filter under the uniform acceleration motion model and the second determinant value of the second observation error covariance matrix of the second sub-filter under the uniform turning motion model are obtained. The ratio of the first determinant value and the second determinant value is compared to obtain the target determinant value with the larger ratio. The motion model corresponding to the target determinant value is used as a reference model. A unified association gate for the uniform acceleration motion model and the uniform turning motion model is established based on the reference model. The motion model is either the uniform acceleration motion model or the uniform turning motion model. Through the association gate, valid point cloud data and first point cloud data corresponding to the valid point cloud data are obtained, and an confirmation matrix is ​​established based on the association relationship between the valid point cloud data and the first point cloud data. Based on the confirmation matrix, joint probability data association is performed to obtain the target motion state and target covariance matrix of each tracked target.

2. The multi-target tracking method for millimeter-wave radar as described in claim 1, characterized in that, Before the step of obtaining the first determinant value of the first observation error covariance matrix of the first sub-filter under the uniformly accelerated motion model, the method includes: Multiple sliding windows are set up to store multiple radar signals output by the millimeter-wave radar in the corresponding sliding windows. When a sliding window operation is detected to be completed, the multiple radar signals corresponding to the currently completed sliding window operation are extracted to obtain multiple frames of target tracking data. Density-based clustering is performed on the multi-frame tracking target data to obtain a cloud of points to be matched.

3. The multi-target tracking method for millimeter-wave radar as described in claim 2, characterized in that, After the step of performing density-based clustering processing on the multi-frame tracking target data to obtain the cloud of points to be matched, the following steps are included: Initialize each of the target trackers; The motion states corresponding to the uniform acceleration motion model and the uniform turning motion model are respectively input into the initialized target tracker for interaction, so as to obtain the weighted total estimated state and weighted total estimated state covariance matrix of the uniform acceleration motion estimation state and the uniform turning motion estimation state of the first sub-filter, and the weighted total estimated state and weighted total estimated state covariance of the uniform acceleration motion estimation state and the uniform turning motion estimation state of the second sub-filter. The target trajectory corresponding to each target point cloud data in the target point cloud set of the previous moment is traversed. Under the uniform acceleration motion model and the uniform turning motion model, the first sub-filter and the second sub-filter corresponding to each target trajectory are predicted and calculated respectively. The predicted motion state value and the predicted state covariance matrix value at the current moment are calculated to obtain the predicted point cloud set.

4. The multi-target tracking method for millimeter-wave radar as described in claim 3, characterized in that, The step of establishing a unified association gate for the uniform acceleration motion model and the uniform turning motion model based on the reference model includes: Each point cloud data to be matched is traversed, and based on the traversed point cloud data to be matched, the predicted point cloud data corresponding to the point cloud data to be matched is traversed to obtain the target predicted point cloud data corresponding to each point cloud data to be matched. Based on the reference model, the association gate is established on each of the target prediction point cloud data.

5. The multi-target tracking method for millimeter-wave radar as described in claim 4, characterized in that, The step of obtaining valid point cloud data through the association gate and establishing an confirmation matrix based on the association relationship between the valid point cloud data and the first point cloud data includes: The association gate is used to determine whether the target point cloud data falls within the association range of the association gate. If the target point cloud data falls within the associated range, the point cloud data to be matched corresponding to the target point cloud data falling within the associated range will be extracted as the valid point cloud data. An initial confirmation matrix is ​​established based on the correlation between the effective point cloud data and the first point cloud data; Obtain multiple predicted point cloud indices associated with each valid point cloud data in the initial confirmation matrix, group valid point cloud data with the same predicted point cloud index into a group, and obtain multiple groups, wherein each group is a confirmation matrix.

6. The multi-target tracking method for millimeter-wave radar as described in claim 5, characterized in that, The step of performing joint probability data association based on the confirmation matrix to obtain the target motion state and target covariance matrix of each tracked target includes: The confirmation matrix is ​​traversed and split to obtain multiple group confirmation matrices. Then, each group confirmation matrix is ​​split to obtain an interconnection matrix, where each interconnection matrix represents a feasible joint event. The probability of the feasible joint event is calculated based on the interconnection matrix to obtain first probability data, and the probability that the valid point cloud data included in the interconnection matrix belongs to the first point cloud data is calculated to obtain second probability data. Based on the comparison relationship between the second probability data and the preset probability threshold, if the second probability data is greater than the preset probability threshold, the matching status of the valid point cloud data corresponding to the second probability data is marked as a correct matching status. The first point cloud data in the correct matching status is determined as the first point cloud data to be calculated, and the valid point cloud data in the correct matching status is determined as the second point cloud data to be calculated. Calculate the Kalman gain and innovation vector of the first point cloud data to be calculated and the second point cloud data to be calculated under the uniform acceleration motion model and the uniform turning motion model, respectively. Based on the Kalman gain, the innovation vector and the first point cloud data to be calculated, update the motion state prediction value and the state covariance matrix prediction value to obtain the next motion state update prediction value and the next state covariance matrix update prediction value. The model matching probability at the current moment is calculated using the maximum likelihood function; Based on the model matching probability, the predicted value of the next motion state update and the predicted value of the next state covariance matrix update are fused to obtain the target motion state and target covariance matrix of the tracked target.

7. The multi-target tracking method for millimeter-wave radar as described in claim 6, characterized in that, After the step of marking the matching status of the valid point cloud data corresponding to the second probability data as a correct matching status if the second probability data is greater than the preset probability threshold, the following steps are included: Valid point cloud data other than the second point cloud data to be calculated are identified as point cloud data to be judged with an incorrect matching status; Determine whether the point cloud data to be judged meets the preset creation conditions; If the point cloud data to be judged meets the preset creation conditions, then the point cloud data to be judged is determined as the first next target point cloud data in the next target tracking process.

8. The multi-target tracking method for millimeter-wave radar as described in claim 7, characterized in that, After the steps of obtaining the target motion state and target covariance matrix of each tracked target, the following steps are included: The track and point cloud attributes corresponding to each first point cloud data are traversed. First point cloud data with a track status of dead track are extracted from each track, and first point cloud data with an attribute status of abnormal attribute are extracted from each point cloud attribute. Then, the first point cloud data with a track status of dead track and / or the first point cloud data with an attribute status of abnormal attribute are removed to obtain the second next target point cloud data. The first next target point cloud data and the second next target point cloud data are input into the next target tracking process to perform calculations for each of the tracked targets.

9. A multi-target tracking device for millimeter-wave radar, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multi-target tracking method for millimeter-wave radar as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the multi-target tracking method of millimeter-wave radar as described in any one of claims 1 to 8.

Images