A vehicle target data association method, system, device and medium based on trajectory state layering

Abstract

The application discloses a vehicle target data association method and system based on track state layering, equipment and medium, relates to the field of multi-target tracking and millimeter wave radar perception technology. The method comprises the following steps: dividing the target track into active track and pending track according to the number of track continuous updates; for the active track, constructing a Gaussian mixture model based on target extension parameters for matching, and combining a global nearest neighbor strategy to complete small vehicle supplementary association; for the pending track, using a density clustering algorithm on the remaining unassociated measurement points to obtain a clustering cluster, constructing a joint association cost function based on the distance between the clustering cluster center and the predicted position of the pending track and the shape difference measure, and completing the data association of the pending track based on the minimum value of the cost function. The application can effectively suppress the large vehicle track splitting and small vehicle track breaking phenomenon, significantly reduce the misassociation rate, and form stable vehicle target tracks, aiming at the characteristics of point cloud sparsity and noise in dense traffic scenes.

Description

Technical Field

[0001] This invention relates to the field of multi-target tracking and millimeter-wave radar sensing technology, specifically to a method, system, device, and medium for associating vehicle target data based on trajectory state hierarchies. Background Technology

[0002] The statements in this section are provided only as background information in connection with this disclosure and may not constitute prior art.

[0003] With the rapid development of intelligent transportation systems and autonomous driving technologies, millimeter-wave radar has been widely applied in the field of road traffic target perception due to its advantages such as all-weather operation, strong anti-interference capabilities, and high accuracy in distance and speed measurement. In vehicle target tracking, data association is a crucial link connecting target detection and state estimation. Its main function is to establish the correspondence between current measurements and historical tracks, thereby continuously maintaining the target's identity and updating its motion state. The accuracy of the data association results directly affects the stability and reliability of the target tracking system.

[0004] Existing vehicle tracking systems typically employ methods such as Nearest Neighbor (NN), Probabilistic Data Association (PDA), and Joint Probabilistic Data Association (JPDA) to match targets with measurements. For millimeter-wave radar point cloud data, a "clustering-tracking" approach is commonly used in engineering. This involves first using clustering algorithms to divide the point cloud into candidate targets, and then combining this with state estimation methods such as Kalman filtering to achieve target tracking. While these methods achieve good tracking results in scenarios with a small number of targets and large target intervals, in dense traffic scenarios, the sparsity, high noise levels, and limited angular resolution of millimeter-wave radar measurement point clouds can easily lead to problems such as target measurement aliasing, false associations, and trajectory splitting, thus affecting tracking performance.

[0005] In engineering applications, given the sparse point clouds and high noise and clutter content of millimeter-wave radar, researchers generally adopt a "clustering-tracking" processing framework. This involves first using clustering algorithms to extract candidate targets from the original point cloud, then establishing a correspondence between measurements and tracks through data association, and finally updating target status and managing the trajectory. To improve the reliability of data association in complex scenarios, Shamsfakhr et al. proposed a multi-target tracking method based on DBSCAN clustering, Kalman filtering, and Structured Branch Multi-Hypothesis Tracking (SBMHT). This method mitigates the impact of measurement uncertainty on target tracking performance by introducing a multi-hypothesis association mechanism (Shamsfakhr F, Macii D, Palopoli L, et al. A multi-target detection and position tracking algorithm based on mmWave-FMCW radar data[J]. Measurement, 2024, 234: 114797). Meanwhile, to address the issue of unstable clustering results caused by changes in point cloud density with target distance in millimeter-wave radar, Zhang et al. proposed an adaptive point cloud clustering algorithm. This algorithm improves target extraction accuracy by dynamically adjusting clustering parameters, thereby providing more reliable measurement input for subsequent data association (Zhang L, Zhang J, Shi H, et al. Adaptive Point Cloud Clustering Algorithm for Practical Roadside MmWave Radar Systems [C]. 2024 IEEE 99th Vehicular Technology Conference (VTC2024-Spring). IEEE, 2024: 1-5). However, most existing methods rely on fixed parameters or empirical thresholds, making it difficult to adapt to different vehicle scales, different detection distances, and dynamically changing point cloud distribution characteristics simultaneously. Therefore, there is an urgent need to study a data association method that can combine vehicle scale characteristics, has parameter adaptive capabilities, and a hierarchical processing mechanism to achieve stable and continuous tracking of vehicle targets in complex traffic scenarios. Summary of the Invention

[0006] The purpose of this invention is to address the problems of traditional association algorithms for large vehicle trajectories splitting and small vehicle trajectories breaking, as well as the inability to adapt to multi-scale vehicle targets, caused by the sparse measurement point cloud and high noise levels of current millimeter-wave radar in dense traffic scenarios. This invention provides a vehicle target data association method, system, device, and medium based on trajectory state hierarchical partitioning. Based on the trajectory state hierarchical partitioning mechanism, a Gaussian mixture model and a supplementary association strategy for small vehicles are used for active tracks, and a density clustering algorithm and a joint cost function of position and shape are used for undetermined tracks. This achieves the technical effect of effectively suppressing large vehicle trajectory splitting and small vehicle trajectory breaking in complex traffic scenarios, reducing the false association rate, and forming stable and continuous target tracks.

[0007] The technical solution of the present invention is as follows: A vehicle target data association method based on trajectory state hierarchy includes: The measurement points of the current frame are obtained, as well as the number of times the target track has been successfully associated and updated in a continuous manner; based on the comparison between the number of times and the preset active track determination threshold, the corresponding target track is divided into active track and undetermined track. For the active trajectory, the corresponding target extension parameters are obtained, and a Gaussian mixture model is constructed based on the target extension parameters; the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model is calculated, and candidate trajectories are extracted based on the comparison result of the distance and a preset threshold. Calculate the measurement association probability between the measurement point and the candidate track, and complete the matching between the measurement point and the candidate track based on the maximum value of the measurement association probability; After completing the matching based on the Gaussian mixture model, the small vehicle track is determined from the active track, and the distance between the first remaining measurement point and the center of the small vehicle track is calculated based on the global nearest neighbor strategy to complete the small vehicle supplementary association. The first remaining measurement point is the measurement point in the current frame that is still not associated after the Gaussian mixture model matching. For the undetermined trajectory, a second remaining measurement point is obtained, and a density clustering algorithm is used to cluster the unprocessed measurement points in the second remaining measurement point to obtain a cluster. The center and shape of the cluster are determined. The second remaining measurement point is the measurement point that is still not associated after the small vehicle supplementary association. Obtain the predicted position and shape of the undetermined trajectory; construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric. The data association between the undetermined trajectory and the cluster is completed based on the minimum value of the joint association cost function.

[0008] Furthermore, based on the comparison result between the number of occurrences and the preset active track determination threshold, the corresponding target track is divided into active tracks and pending tracks, including: The number of times the target track has been successfully associated and its status updated consecutively. ,in Indicates the first A flight path; Determine the number of times Is it greater than or equal to the preset active track determination threshold? ; If so, the corresponding target track will be classified as an active track; If not, the corresponding target track will be classified as a track to be determined.

[0009] Furthermore, the target extension parameters include the four corner points of the vehicle target and the vehicle's geometric center; the construction of a Gaussian mixture model based on the target extension parameters includes: The Gaussian mixture model is constructed using five Gaussian components, the means of which are denoted as follows: ,in The index of the Gaussian component, with a value from 1 to 5; Let the four corner points of the vehicle target be, in clockwise order, as follows: The geometric center of the vehicle is Then the set of means of each Gaussian component is defined as: .

[0010] Further, the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model is calculated. Based on the comparison result of the distance and a preset threshold, candidate tracks are extracted, including: For the Any measurement point within the frame ,in Indicate the serial number of the measurement point, and calculate its correlation with the active track. In the Mean Gaussian component of the frame The Euclidean distance between the measurement points is calculated, and the minimum value is selected as the correlation distance between the measurement point and the corresponding active track. The calculation formula is:

[0011] in: Represents the Euclidean distance operator; The correlation distance If the active track is compared with a preset threshold and the threshold constraint is met, the corresponding active track is extracted as a candidate track.

[0012] Furthermore, a density-based clustering algorithm is used to cluster the remaining measurement points in the second set of measurement points to obtain clusters, including: Based on the target shape parameters estimated by the random matrix model, small vehicle undetermined tracks are selected from the undetermined tracks. Obtain the predicted center position of the undetermined trajectory of the small vehicle, and calculate the Euclidean distance between the second remaining measurement point and the corresponding predicted center position; Based on the Euclidean distance and using the global nearest neighbor algorithm, the preliminary measurement and allocation of the undetermined trajectory of the small vehicle is completed. Measurement points that are still not associated after the initial measurement allocation are taken as the measurement points to be processed and input into the density clustering algorithm for clustering to obtain the clusters.

[0013] Furthermore, the density clustering algorithm is specifically the density-based noise-applied spatial clustering algorithm DBSCAN.

[0014] Furthermore, the formula for calculating the joint association cost function is as follows:

[0015] in: The value of the joint association cost function; Indicates a pending flight path Predicted location With clusters center coordinates The Euclidean distance between them; , These represent undetermined flight paths. The estimated length and width of the shape; , Representing clusters The estimated length and width of the obtained shape; These are the preset weight parameters.

[0016] This invention also proposes a vehicle target data association system based on trajectory state hierarchy, which can implement the above method, including: The state division module is used to obtain the measurement points of the current frame and the number of times the target track has been successfully associated and completed the state update. Based on the comparison result of the number of times and the preset active track determination threshold, the corresponding target track is divided into active track and undetermined track. The active trajectory association module is used to obtain the corresponding target extended parameters for the active trajectory, construct a Gaussian mixture model based on the target extended parameters, calculate the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model, extract candidate trajectories based on the comparison result of the distance and a preset threshold, calculate the measurement association probability between the measurement point and the candidate trajectory, and complete the matching between the measurement point and the candidate trajectory based on the maximum value of the measurement association probability. The supplementary association module is used to determine the small vehicle track from the active track after completing the matching based on the Gaussian mixture model, and calculate the distance between the first remaining measurement point and the center of the small vehicle track based on the global nearest neighbor strategy to complete the supplementary association of the small vehicle. The first remaining measurement point is the measurement point in the current frame that is still not associated after the matching based on the Gaussian mixture model. The clustering processing module is used to obtain the second remaining measurement points for the undetermined trajectory, and to use a density clustering algorithm to cluster the unprocessed measurement points in the second remaining measurement points to obtain clusters, and to determine the center and shape of the clusters, wherein the second remaining measurement points are measurement points that are still not associated after the small vehicle supplementary association; The undetermined trajectory association module is used to obtain the predicted position and shape of the undetermined trajectory; to construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric; and to complete the data association between the undetermined trajectory and the cluster based on the minimum value of the joint association cost function.

[0017] The present invention also proposes an electronic device, comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the instructions stored in the memory to perform the method described above.

[0018] The present invention also proposes a computer-readable storage medium for storing instructions that, when executed, cause the method described above to be implemented.

[0019] Compared with existing technologies, the advantages of this invention are: 1. Effectively suppresses trajectory splitting of large vehicles. This invention breaks through the limitations of traditional single-center-point association models. For convergent active trajectories, a Gaussian mixture model with 5 components is constructed based on the vehicle's extended parameters (four corner points and the center). This model is like a probability net that fits the vehicle's contour, effectively capturing measurement points scattered around the vehicle's edges, greatly reducing the probability of large vehicles splitting in the radar field of view.

[0020] 2. Completely solves the problem of fragmented trajectories for small vehicles. Addressing the issue of sparse point clouds and easy omissions in detecting small vehicles, this invention, after matching active trajectory models, separately designs a supplementary association mechanism for small vehicles based on a global nearest neighbor strategy. This "fallback" design effectively absorbs discrete measurement points, ensuring the continuity of the vehicle's trajectory even in environments with very few point clouds.

[0021] 3. Enhance the robustness of initial / undetermined track construction. For undetermined tracks in the initialization or state establishment phase, this invention does not blindly associate them, but introduces a joint association cost function of shape and position. By comprehensively considering the center distance and length-width-shape differences between clusters and predicted tracks, the problem of mis-clustering of multiple vehicle targets in dense scenes is completely avoided, greatly reducing the occurrence rate of false associations. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 A flowchart illustrating a vehicle target data association method based on trajectory state hierarchies provided in an embodiment of the present invention; Figure 2 This is a simulation scene diagram provided for an embodiment of the present invention; Figure 3 The image corresponding to the 6th frame provided in this embodiment of the invention; Figure 4 The tracking result diagram for the 6th frame provided in this embodiment of the invention; Figure 5 The Gaussian mixture correlation model diagram for the 6th frame with active track id=9 provided in this embodiment of the invention; Figure 6 This is an updated track map after associating the undetermined track id=13, as provided in an embodiment of the present invention. Figure 7 The image corresponding to frame 20 provided in this embodiment of the invention; Figure 8This is a tracking result diagram of the 20th frame provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0026] Example 1 In scenarios with dense intelligent traffic flow, due to the sparsity, high noise, and limited angular resolution of millimeter-wave radar measurement point clouds, problems such as overlapping measurements of large vehicle targets leading to trajectory splitting and the trajectory breakage of small vehicle targets due to a lack of measurement points are very likely to occur.

[0027] Generally, target measurement generation conforms to an extended target measurement model based on the target's true state. To reflect the radar's observation characteristics of "dense point clouds at close range and sparse point clouds at long range," the number of target measurement points typically follows a Poisson distribution, with its mean parameter... Related to the target size and the distance from the target to the radar, it can be expressed as:

[0028] in, The distance between the target and the radar. These are empirical parameters. is the scale factor.

[0029] Based on the aforementioned radar point cloud characteristics, this embodiment provides a vehicle target data association method based on trajectory state layering, the processing flow of which is as follows: Figure 1 As shown, the method includes the following specific steps: Step S1: Trajectory state division; specifically: The system acquires the measurement points of the current frame and the number of times the target track has been successfully associated and updated consecutively. Based on a comparison of these counts with a preset active track determination threshold, the corresponding target track is divided into active tracks and pending tracks. Specifically, this includes: acquiring the number of times the target track has been successfully associated and updated consecutively. ,in Indicates the first A flight path; determine the number of times. Is it greater than or equal to the preset active track determination threshold? (In this embodiment, empirical parameters are preferably set) If yes, the corresponding target track is classified as an active track; otherwise, it is classified as a pending track.

[0030] In this embodiment, it should be noted that in engineering practice, target tracks are typically maintained using state estimation algorithms based on Kalman filtering. Active tracks, having undergone continuous measurement updates across multiple frames, have largely converged their motion state parameters and extended shape parameters, resulting in a significant reduction in the corresponding state covariance matrix, thus exhibiting high reliability in state estimation. Conversely, undetermined tracks are usually in the initialization or state establishment phase, where their motion state and extended shape parameters have not yet fully converged, leading to significant uncertainty in state estimation. Therefore, a more robust data association strategy is required. Based on this physical principle, this invention innovatively proposes a "layered processing" architecture to prevent unstable tracks from "stealing" measurement points from high-quality tracks.

[0031] Step S2: Active track association; specifically: For the active flight path, the corresponding target extension parameters (including the four corner points of the vehicle target and the vehicle's geometric center) are obtained, and a Gaussian mixture model is constructed based on these parameters. Specifically, the Gaussian mixture model is constructed using five Gaussian components, and the means of these five Gaussian components are denoted as... ,in The index represents the Gaussian component, with values ​​ranging from 1 to 5. Let the four corner points of the vehicle target be arranged clockwise as follows: The geometric center of the vehicle is Then the set of means of each Gaussian component is defined as:

[0032] Further, the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model is calculated. Based on the comparison result of the distance and a preset threshold, candidate tracks are extracted. Specifically, this includes: For the Any measurement point within the frame ,in Indicate the serial number of the measurement point, and calculate its correlation with the active track. In the Mean Gaussian component of the frame The minimum Euclidean distance between them is selected as the association distance. ,in, This represents the Euclidean distance operator. If the associated distance satisfies a preset threshold constraint, it is extracted as a candidate track.

[0033] Subsequently, the measurement association probability between the measurement point and the candidate track is calculated, and matching is performed based on the maximum probability. Specifically, the association probability model between the measurement point and the target can be defined as:

[0034] in: This indicates that the measurement point and the target are at the [number]th ...]. Frame measurement correlation probability; Indicates the target is in the first place. Frame motion state parameters; Indicates the target is in the first place. Extended shape parameters of the frame; Let be the covariance matrix of the current target motion; For weighted parameters (e.g., preferably 0.2); Measurement point; The probability density function is Gaussian.

[0035] Large vehicles exhibit significant spatial expansion within the radar field of view. Traditional single-center-point models are prone to misclassifying measurement points at the front and rear bumpers of large vehicles as new targets due to their excessive distance from the center, resulting in "track splitting." This invention creatively utilizes converged expansion parameters to establish five Gaussian components at the geometric center of the vehicle and the midpoints of its four edges. This model acts like a "probability net conforming to the vehicle's contours," effectively attracting measurement points scattered along the vehicle's edges and suppressing track splitting in large vehicles.

[0036] Step S3: Supplement the association for small vehicles; specifically: After completing the matching based on the Gaussian mixture model, the small vehicle track is determined from the active tracks, and the distance between the first remaining measurement point and the center of the small vehicle track is calculated based on the Global Nearest Neighbor (GNN) strategy to complete the supplementary association of the small vehicle. The first remaining measurement point is a measurement point in the current frame that has not yet been associated after matching using the Gaussian mixture model.

[0037] In this embodiment, it should be noted that, due to radar resolution limitations, the number of measurement points corresponding to small vehicle targets (such as cars or motorcycles) is extremely small. If only the strict probability threshold of the Gaussian mixture model is relied upon, sporadic measurement points are easily missed. To address this, the present invention specifically designs a "fallback mechanism": after the large cycle of active tracks ends, a GNN is specifically activated to supplement the association for small vehicle targets, using their center position to "attract" the first remaining measurement points, thereby ensuring that the small vehicle track does not break under sparse measurement conditions.

[0038] Step S4: Preprocessing for clustering undetermined tracks; specifically: For the undetermined trajectory, a second set of remaining measurement points (i.e., measurement points that are still not associated after the small vehicle supplementary association) are obtained, and a density clustering algorithm (preferably DBSCAN) is used to cluster the unprocessed measurement points in the second set of remaining measurement points.

[0039] Before clustering, the following filtering process is included to prevent false clustering: 1) Based on the target shape parameters estimated by the random matrix model, select small vehicle undetermined tracks from the undetermined tracks (for example, based on empirical parameters, select elliptical targets with a semi-axis length of less than 2m and a semi-axis width of less than 0.8m as undetermined tracks for small vehicles). 2) Obtain the predicted center position of the undetermined trajectory of the small vehicle, and calculate the Euclidean distance between the second remaining measurement point and the predicted center position; 3) Based on Euclidean distance and using the global nearest neighbor algorithm, complete the preliminary measurement and allocation of the undetermined trajectories of small vehicles (e.g., set the distance threshold to 1.2m). 4) Measurement points that are still not associated after the initial measurement allocation are taken as the measurement points to be processed, input into the DBSCAN algorithm to obtain clusters, and determine the center and shape of the clusters.

[0040] In this embodiment, it should be noted that if the undetermined trajectory is directly included in the global clustering association because it has not converged sufficiently, it is prone to bias. Especially in scenarios with dense spatial distribution of non-motorized vehicles / cars, the DBSCAN algorithm can easily misclassify (cluster) multiple adjacent small targets into a large cluster. This invention uses a random matrix model to estimate the shape in advance and preemptively allocates the discrete measurement points belonging to cars, making the measurement points available for the DBSCAN algorithm clean and fundamentally preventing the misclustering of dense small targets.

[0041] Step S5: Allocation of joint association costs for pending paths; specifically: Obtain the predicted position and shape of the undetermined trajectory; construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost, and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric. The calculation formula is as follows:

[0042] in: The value of the joint association cost function; Indicates a pending flight path Predicted location With clusters center coordinates The Euclidean distance between them; , These represent undetermined flight paths. The estimated length and width of the shape; , Representing clusters The estimated length and width of the obtained shape; These are the preset weight parameters.

[0043] Finally, the data association between the undetermined trajectory and the cluster is completed based on the minimum value of the joint association cost function.

[0044] In this embodiment, it should be noted that for undetermined tracks that are not yet stable, simply comparing center distances results in a large error. This invention creatively introduces "shape similarity" into the cost function. The joint cost is minimized only when the centers of clusters are close and their shapes (length and width) are highly similar. This significantly improves the robustness of associations in complex traffic scenarios. Through the aforementioned hierarchical association strategy, this invention can effectively suppress the splitting of large vehicle targets and the misassociation of small vehicles in scenarios where both large and small vehicles coexist, forming stable and continuous tracks.

[0045] Example 2 Based on the method described in Embodiment 1, this embodiment provides a specific MATLAB simulation verification example to further illustrate the specific implementation process and beneficial effects of the present invention.

[0046] 1. Simulation Scene and Measurement Generation Parameter Configuration Simulation of road scenes, such as Figure 2 As shown, the measurement generation adopts an extended target measurement model based on the target's true state. For each vehicle target in a frame, scattering points are randomly generated in the target's local coordinate system according to its true position, size, and orientation parameters, and then mapped to the global coordinate system through rotation transformation to simulate the vehicle's spatial expansion characteristics.

[0047] The number of target measurement points follows a Poisson distribution, and its mean parameter is related to the target size and the distance from the target to the radar, expressed as: .in, The distance between the target and the radar. These are empirical parameters. The scaling factor is used. This model can accurately reflect the radar observation characteristics of "dense point clouds at close range and sparse point clouds at long range".

[0048] The specific simulation parameter settings for this embodiment are shown in Table 1: Table 1 Simulation Parameters

[0049] 2. Simulation process of trajectory state division Select the 6th frame of the simulation scene for display (e.g.) Figure 3 and Figure 4 (As shown). There are 18 targets in this scenario, one of which failed to be tracked due to the sparse measurement points in the current frame, resulting in only 17 valid tracks. An empirical threshold is used based on the number of track updates. Distinguish between flight paths: For consecutive successful updates The tracks are identified as active tracks. The active track IDs at this time are 3, 6, 7, 8, 9, and 10. Active tracks are associated with these tracks first. For the remaining consecutive successful update counts The track is classified as a pending track, and the pending track IDs are 1, 2, 4, 5, 11, 12, 13, 14, 15, 16, and 17.

[0050] 3. Simulation process of active trajectory association (taking id=9 as an example) For meeting the threshold The active tracks are analyzed, and a Gaussian mixture model is constructed based on their extended parameters. For example, for the active track with vehicle ID 9, a model like this can be constructed. Figure 5 The Gaussian mixture model is shown.

[0051] Calculate the distances between surrounding measurement points and the Gaussian component, and select the minimum value to determine whether this point is a measurement to be assigned to the trajectory. If the minimum value meets the set empirical threshold (2m), calculate the association probability of this point under the Gaussian mixture model. For the track with id 9, the relevant data calculated from four surrounding measurement points are shown in Table 2. Table 2. Relevant data calculated from the four surrounding measurement points.

[0052] By calculating the probability of each track and measurement point, a correlation probability matrix is ​​obtained. In this matrix, rows represent flight paths and columns represent measurement points. In this embodiment, the above four points are selected and compared with the probability matrices for id 9 and id 3, as shown in Table 3: Table 3. Allocation Comparison Results

[0053] According to the maximum probability allocation principle in the table above, the points (-0.51, -7.78)m and (2.02, -7.86)m were successfully assigned to the target with id 9.

[0054] 4. Simulation process of undetermined trajectory association (taking id=13 as an example) After active trajectory model matching and small vehicle GNN supplementary association, pending trajectory association is performed on the remaining unassociated measurement points.

[0055] First, vehicle tracks are selected based on estimated target expansion parameters. For example, based on empirical parameters, targets with an elliptical target semi-axis length within 2m and a semi-axis width within 0.8m are selected as potential vehicle tracks. Based on the predicted center position, the Euclidean distance between each measurement point and the track is calculated, with a distance threshold of 1.2m set. A GNN association algorithm is preferentially used for initial association.

[0056] For measurement points that remain unassociated after initial association, the DBSCAN algorithm is used for clustering. In this embodiment, one cluster is obtained, and the coordinates of the points contained in this cluster are (1.6810, 3.7516) m, (1.9296, 3.075) m, (2.4021, 4.2961) m, and (0.9300, 3.8017) m.

[0057] The shape parameters of this cluster were extracted and calculated as follows: the semi-axis length is 1.2245m, the semi-axis width is 0.7145m, and the center position is located at (1.7357, 3.7892)m.

[0058] Based on the joint association cost function formula, the weighted association cost (joint difference in position and shape) between this cluster and the undetermined track with id 13 is minimized, with a distance value of 1.6900m. Therefore, this cluster is successfully associated with the undetermined track with id 13. The updated track information after association is as follows. Figure 6 As shown.

[0059] 5. Final tracking effect verification Using the method of this embodiment, the scene was continued to be stably tracked, and the tracking result was obtained after 14 frames (i.e., the 20th frame). Figure 7 and Figure 8 As shown.

[0060] The simulation results show that, with Figure 5 Compared to the initial results in frame 6, the previously sparsely measured targets have now obtained sufficient measurement support and completed correct association, allowing their corresponding tracks to be stably established. Simultaneously, the measurement and track matching relationships for each target remain correct, effectively suppressing the target splitting phenomenon that is prone to occur with large vehicles and avoiding track breakage for small vehicles. This verifies that the data association method provided by this invention has extremely high association stability and tracking reliability in both the target initialization stage and subsequent tracking processes.

[0061] Example 3 Based on the same inventive concept as the vehicle target data association method based on trajectory state hierarchy in Embodiment 1 above, this embodiment provides a vehicle target data association system based on trajectory state hierarchy. Since the principle by which this system solves the problem is similar to that of the method embodiment described above, the implementation of this system can be referred to the implementation of the method, and repeated details will not be elaborated further.

[0062] Specifically, this embodiment provides a vehicle target data association system based on trajectory state hierarchies, including: The state division module is used to obtain the measurement points of the current frame and the number of times the target track has been successfully associated and completed the state update. Based on the comparison result of the number of times and the preset active track determination threshold, the corresponding target track is divided into active track and undetermined track. The active trajectory association module is used to obtain the corresponding target extended parameters for the active trajectory, construct a Gaussian mixture model based on the target extended parameters, calculate the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model, extract candidate trajectories based on the comparison result of the distance and a preset threshold, calculate the measurement association probability between the measurement point and the candidate trajectory, and complete the matching between the measurement point and the candidate trajectory based on the maximum value of the measurement association probability. The supplementary association module is used to determine the small vehicle track from the active track after completing the matching based on the Gaussian mixture model, and calculate the distance between the first remaining measurement point and the center of the small vehicle track based on the global nearest neighbor strategy to complete the supplementary association of the small vehicle. The first remaining measurement point is the measurement point in the current frame that is still not associated after the matching based on the Gaussian mixture model. The clustering processing module is used to obtain the second remaining measurement points for the undetermined trajectory, and to use a density clustering algorithm to cluster the unprocessed measurement points in the second remaining measurement points to obtain clusters, and to determine the center and shape of the clusters, wherein the second remaining measurement points are measurement points that are still not associated after the small vehicle supplementary association; The undetermined trajectory association module is used to obtain the predicted position and shape of the undetermined trajectory; to construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric; and to complete the data association between the undetermined trajectory and the cluster based on the minimum value of the joint association cost function.

[0063] It should be noted that the system described in this embodiment is a virtual functional module (e.g., a pure software program module) corresponding to the method in Embodiment 1. The specific technical details of each module's execution and the resulting effects are completely consistent with Embodiments 1 and 2, and will not be elaborated further here.

[0064] Based on the same technical concept, embodiments of the present invention also provide an electronic device that can implement the vehicle target data association method based on trajectory state hierarchies provided in the above embodiments of the present invention. In one embodiment, the electronic device can be a server, a terminal device, or other electronic devices. Figure 9 As shown, the electronic device may include: At least one processor and a memory connected to the at least one processor. In this embodiment of the invention, the specific connection medium between the processor and the memory is not limited. Figure 9 The example used is the connection between the processor and memory via a bus. The bus... Figure 9The connections between other components are indicated by thick lines and are for illustrative purposes only, not as limiting information. Buses can be categorized into address buses, data buses, control buses, etc., but for ease of representation, [the specific bus type is not shown here]. Figure 9 The processor is represented by a single thick line, but this does not imply that there is only one bus or one type of bus. Alternatively, a processor can also be called a controller; there are no restrictions on the name.

[0065] In this embodiment of the invention, the memory stores instructions executable by at least one processor. By executing the instructions stored in the memory, the at least one processor can perform the vehicle target data association method based on trajectory state hierarchies discussed above. The processor can implement... Figure 9 The functions of each module in the device shown.

[0066] The processor is the control center of the device. It can connect to various parts of the control device through various interfaces and lines. By running or executing instructions stored in memory and calling data stored in memory, it can monitor the device's various functions and process data, thereby enabling overall monitoring of the device.

[0067] In an alternative design, the processor may include one or more processing units. The processor may integrate an application processor and a modem processor, wherein the application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. In some embodiments, the processor and memory may be implemented on the same chip; in some embodiments, they may also be implemented separately on separate chips.

[0068] The processor can be a general-purpose processor, such as a CPU, digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the vehicle target data association method based on trajectory state layering disclosed in the embodiments of this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0069] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory can include at least one type of storage medium, such as flash memory, hard disk, multimedia cards, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), and electrically erasable programmable read-only memory (EPROM). Only memory (EEPROM), magnetic storage, magnetic disks, optical disks, etc. A memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer, but is not limited thereto. The memory in embodiments of this invention can also be a circuit or any other device capable of performing storage functions for storing program instructions and / or data.

[0070] By designing and programming the processor, the code corresponding to the vehicle target data association method based on trajectory state layering described in the foregoing embodiments can be embedded into the chip, enabling the chip to execute the steps of the method described in the foregoing embodiments during runtime. How to design and program the processor is a technique well-known to those skilled in the art and will not be elaborated upon here.

[0071] Based on the same inventive concept, embodiments of the present invention also provide a storage medium storing computer instructions that, when executed on a computer, cause the computer to perform a vehicle target data association method based on trajectory state hierarchies as described above.

[0072] In some alternative embodiments, the present invention also provides a method for associating vehicle target data based on trajectory state hierarchy, which can also be implemented as a program product including program code. When the program product is run on a device, the program code is used to cause the control device to perform the steps in the method for associating vehicle target data based on trajectory state hierarchy according to various exemplary embodiments of the present invention as described above.

[0073] It should be noted that although several units or sub-units of the apparatus have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units. Furthermore, although the operation of the method of the invention is described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0074] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can be implemented in one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs) containing computer-usable program code. The form of a computer program product implemented on ROM, optical memory, etc.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a server, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] Program code for performing the operations of this invention can be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0077] In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

[0081] This background section is provided to generally present the context of the invention. The work of the currently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither expressly nor impliedly acknowledged as prior art to the invention.

Claims

1. A method for associating vehicle target data based on trajectory state hierarchies, characterized in that, include: Obtain the measurement points of the current frame, and the number of times the target track has been successfully associated and updated in a continuous manner; Based on the comparison results between the number of times and the preset active track determination threshold, the corresponding target track is divided into active track and undetermined track. For the active trajectory, obtain the corresponding target extension parameters, and construct a Gaussian mixture model based on the target extension parameters; Calculate the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model, and extract candidate tracks based on the comparison result of the distance and a preset threshold; Calculate the measurement association probability between the measurement point and the candidate track, and complete the matching between the measurement point and the candidate track based on the maximum value of the measurement association probability; After completing the matching based on the Gaussian mixture model, the small vehicle track is determined from the active track, and the distance between the first remaining measurement point and the center of the small vehicle track is calculated based on the global nearest neighbor strategy to complete the small vehicle supplementary association. The first remaining measurement point is the measurement point in the current frame that is still not associated after the Gaussian mixture model matching. For the undetermined trajectory, a second remaining measurement point is obtained, and a density clustering algorithm is used to cluster the unprocessed measurement points in the second remaining measurement point to obtain a cluster. The center and shape of the cluster are determined. The second remaining measurement point is the measurement point that is still not associated after the small vehicle supplementary association. Obtain the predicted position and shape of the undetermined trajectory; construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric. The data association between the undetermined trajectory and the cluster is completed based on the minimum value of the joint association cost function.

2. The vehicle target data association method based on trajectory state hierarchical as described in claim 1, characterized in that, Based on the comparison result between the number of occurrences and the preset active track determination threshold, the corresponding target track is divided into active tracks and pending tracks, including: The number of times the target track has been successfully associated and its status updated consecutively. ,in Indicates the first A flight path; Determine the number of times Is it greater than or equal to the preset active track determination threshold? ; If so, the corresponding target track will be classified as an active track; If not, the corresponding target track will be classified as a track to be determined.

3. The vehicle target data association method based on trajectory state hierarchical as described in claim 1, characterized in that, The target extension parameters include the four corner points of the vehicle target and the vehicle's geometric center; the construction of a Gaussian mixture model based on the target extension parameters includes: The Gaussian mixture model is constructed using five Gaussian components, the means of which are denoted as follows: ,in The index represents the Gaussian component, with values ​​ranging from 1 to 5; Let the four corner points of the vehicle target be, in clockwise order, as follows: The geometric center of the vehicle is Then the set of means of each Gaussian component is defined as: 。 4. The vehicle target data association method based on trajectory state hierarchical as described in claim 3, characterized in that, Calculate the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model. Based on the comparison result of the distance and a preset threshold, extract candidate tracks, including: For the Any measurement point within the frame ,in Indicate the serial number of the measurement point, and calculate its correlation with the active track. In the Gaussian component mean of the frame The Euclidean distance between the measurement points is calculated, and the minimum value is selected as the correlation distance between the measurement point and the corresponding active track. The calculation formula is: in: Represents the Euclidean distance operator; The correlation distance If the active track is compared with a preset threshold and the threshold constraint is met, the corresponding active track is extracted as a candidate track.

5. The vehicle target data association method based on trajectory state hierarchical as described in claim 1, characterized in that, Density clustering algorithm is used to cluster the measurement points to be processed in the second remaining measurement points to obtain clusters, including: Based on the target shape parameters estimated by the random matrix model, small vehicle undetermined tracks are selected from the undetermined tracks. Obtain the predicted center position of the undetermined trajectory of the small vehicle, and calculate the Euclidean distance between the second remaining measurement point and the corresponding predicted center position; Based on the Euclidean distance and using the global nearest neighbor algorithm, the preliminary measurement and allocation of the undetermined trajectory of the small vehicle is completed. Measurement points that are still not associated after the initial measurement allocation are taken as the measurement points to be processed and input into the density clustering algorithm for clustering to obtain the clusters.

6. The vehicle target data association method based on trajectory state hierarchical as described in claim 1, characterized in that, The density clustering algorithm is specifically the density-based noise-applied spatial clustering algorithm DBSCAN.

7. The vehicle target data association method based on trajectory state hierarchical as described in claim 1, characterized in that, The formula for calculating the joint association cost function is as follows: in: The value of the joint association cost function; Indicates a pending flight path Predicted location With clusters center coordinates The Euclidean distance between them; , These represent undetermined flight paths. The estimated length and width of the shape; , Representing clusters The estimated length and width of the obtained shape; These are the preset weight parameters.

8. A vehicle target data association system based on trajectory state hierarchies, characterized in that, Capable of implementing the method according to any one of claims 1-7, comprising: The state division module is used to obtain the measurement points of the current frame and the number of times the target track has been successfully associated and completed the state update. Based on the comparison result of the number of times and the preset active track determination threshold, the corresponding target track is divided into active track and undetermined track. The active trajectory association module is used to obtain the corresponding target extended parameters for the active trajectory, construct a Gaussian mixture model based on the target extended parameters, calculate the distance between the measurement point and the mean of each Gaussian component in the Gaussian mixture model, extract candidate trajectories based on the comparison result of the distance and a preset threshold, calculate the measurement association probability between the measurement point and the candidate trajectory, and complete the matching between the measurement point and the candidate trajectory based on the maximum value of the measurement association probability. The supplementary association module is used to determine the small vehicle track from the active track after completing the matching based on the Gaussian mixture model, and calculate the distance between the first remaining measurement point and the center of the small vehicle track based on the global nearest neighbor strategy to complete the supplementary association of the small vehicle. The first remaining measurement point is the measurement point in the current frame that is still not associated after the matching based on the Gaussian mixture model. The clustering processing module is used to obtain the second remaining measurement points for the undetermined trajectory, and to use a density clustering algorithm to cluster the unprocessed measurement points in the second remaining measurement points to obtain clusters, and to determine the center and shape of the clusters, wherein the second remaining measurement points are measurement points that are still not associated after the small vehicle supplementary association; The undetermined trajectory association module is used to obtain the predicted position and shape of the undetermined trajectory; to construct a joint association cost function using the distance between the center of the cluster and the predicted position of the undetermined trajectory as the position cost and the difference between the shape of the cluster and the shape of the undetermined trajectory as the shape difference metric; and to complete the data association between the undetermined trajectory and the cluster based on the minimum value of the joint association cost function.

9. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which executes the instructions stored in the memory to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store instructions that, when executed, cause the method as described in any one of claims 1-7 to be implemented.

Images