Target tracking method and system based on DBSCAN and Kmeans

Abstract

The invention discloses a target tracking method and system based on DBSCAN and Kmeans, and relates to the technical field of target tracking, and the method comprises the steps: marking all point cloud data sets as unaccessed; all the point cloud data sets are sorted in a descending order according to the signal-to-noise ratio; sequentially selecting a point as a research object; setting a correlation radius according to the speed of the research object, and setting a density threshold according to the signal-to-noise ratio of the research object; carrying out clustering analysis on the research object by adopting a DBSCAN method, and outputting a cluster; performing clustering judgment on the clustering clusters; if not, outputting a clustering cluster; if yes, the number of possible targets is calculated; and adopting a Kmeans algorithm to perform secondary clustering on the DBSCAN cluster, and outputting a target number of new clusters. According to the method, clustering judgment is added after DBSCAN clustering, the same target is prevented from being divided into multiple clusters, the accuracy of a clustering result is improved, and the accuracy of traffic flow statistics is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of target tracking, and in particular to a target tracking method and system based on DBSCAN and Kmeans. Background Technique

[0002] As a common sensor, millimeter-wave radar has gradually been applied to the field of road traffic auxiliary monitoring with its advantages of large detection range, high detection accuracy, and strong anti-interference ability. It can perform real-time tracking of multiple targets, realize functions such as vehicle flow statistics and speeding detection, and provide technical support and data support for intelligent transportation systems. As the carrier frequency of millimeter-wave radar gradually migrates from 24 GHz to 77 GHz and higher frequency bands, the resolution of millimeter-wave radar is getting higher and higher, and the number of scatter points obtained by processing the reflected echoes of the same target also increases. At the same time, in actual application scenarios, multiple targets need to be detected and tracked. Therefore, before tracking the target, how to correctly cluster multiple scatterings of the same target together is an important research direction in the field of millimeter-wave radar today.

[0003] The clustering algorithm is a solution strategy for unsupervised learning cases, which is applicable when the information of classes is unknown, and divides data into clusters according to the similarity of data.

[0004] Different clustering algorithms are applicable to different data scenarios. According to the characteristics of millimeter-wave point cloud radar data, the density-based DBSCAN clustering algorithm is the most common and effective clustering method. Currently, in the field of millimeter-wave point cloud radar clustering, people mainly optimize based on the DBSCAN algorithm to adapt to different scenarios and different targets. In a certain actual traffic scenario, two vehicles are moving forward in parallel. Projecting the traces onto the Cartesian coordinate system, we get a trace distribution map as shown in Figure 1 The red circles are the scatter points of the vehicle marked by the red frame in the actual scenario, and the blue stars are the scatter points of the vehicle marked by the blue frame in the actual scenario. Using the general DBSCAN algorithm will cluster the two vehicles into one cluster, which is likely to cause the problem of losing the tracking target. Moreover, if the association radius and density threshold are set improperly and the same target is divided into multiple clusters, resulting in over-clustering, the target tracking result will be inaccurate. Summary of the Invention

[0005] The purpose of the present invention is to provide a target tracking method and system based on DBSCAN and Kmeans, which solves the problems of target loss and inaccurate target tracking results when multiple vehicles are driving in parallel.

[0006] The present invention is realized through the following technical solutions: In the first aspect, a target tracking method based on DBSCAN and Kmeans provided by an embodiment of the present invention includes: Use a millimeter-wave radar to obtain a point cloud data set of moving targets in a road scene; Mark all point cloud data sets as unvisited; Sort all point cloud data sets in descending order according to the signal-to-noise ratio; Select a point in order as the research object and mark the research object as visited; Set the association radius according to the speed of the research object and set the density threshold according to the signal-to-noise ratio of the research object; Use the DBSCAN clustering method to perform clustering analysis on the research object and output the DBSCAN clustering clusters; Use the prior target lateral dimension as a threshold to perform over-clustering judgment on the DBSCAN clustering clusters; If there is no over-clustering, output the DBSCAN clustering clusters; If there is over-clustering, calculate the possible number of targets; Use the Kmeans algorithm to perform secondary clustering on the DBSCAN clustering clusters and output a new cluster of the number of targets.

[0007] Furthermore, the specific method for setting the association radius according to the speed of the research object includes: The function of the association radius set according to the research object changing with speed is: ; where, Ɛ is the association radius, unit is m, v is the speed of the trace, unit is m / s.

[0008] Furthermore, the specific method for setting the density threshold according to the signal-to-noise ratio of the research object includes: Set the density threshold during clustering according to the signal-to-noise ratio, and the calculation formula for the value of the density threshold is: ; where, MinPts is the density threshold, snr is the signal-to-noise ratio, unit is dB.

[0009] Furthermore, the specific method for using the prior target lateral dimension as a threshold to perform over-clustering judgment on the DBSCAN clustering clusters includes: Perform point trace profile analysis on the points inside the clustering cluster and calculate the vehicle length and vehicle width information; Set the maximum vehicle width threshold and compare the vehicle width with the maximum vehicle width threshold; If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering; If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

[0010] Further, the specific method for performing clustering analysis on the research object using the DBSCAN clustering method includes: Compare the number of objects within the neighborhood of the association radius of the research object with the density threshold; If the number of objects is greater than the density threshold, the research object is a noise point; If the number of objects is less than or equal to the density threshold, the research object is a core point, and the research object and the objects within its associated neighborhood are added to the newly created cluster; Select a point from the newly created cluster. If the point is marked as unvisited, mark it as visited, calculate the number of objects within the neighborhood of the association radius of this point, and compare the number of objects within the neighborhood of the association radius of this point with the density threshold; If the number of objects within the neighborhood of the association radius of this point is greater than the density threshold, this point is a core point, and the objects within its neighborhood are added to the newly created cluster; If the number of objects within the neighborhood of the association radius of this point is less than or equal to the density threshold, this point is a boundary point; Loop through until all points in the newly created cluster are marked as visited.

[0011] In a second aspect, a target tracking system based on DBSCAN and Kmeans provided by an embodiment of the present invention includes: a data acquisition module, a first clustering module, an over-clustering judgment module, and a second clustering module. The data acquisition module uses a millimeter-wave radar to obtain a point cloud data set of moving targets in a road scene. The first clustering module is used to mark all point cloud data sets as unvisited, sort all point cloud data sets in descending order according to the signal-to-noise ratio, select a point as the research object in order, mark the research object as visited, set the association radius according to the speed of the research object, set the density threshold according to the signal-to-noise ratio of the research object, and perform clustering analysis on the research object using the DBSCAN clustering method, and output the DBSCAN clustering clusters. The over-clustering judgment module uses the prior target lateral size as a threshold to perform over-clustering judgment on the DBSCAN clustering clusters. If there is no over-clustering, output the DBSCAN clustering clusters. If there is over-clustering, calculate the possible number of targets. The second clustering module is used to perform secondary clustering on the DBSCAN clustering clusters using the Kmeans algorithm and output a new cluster of the number of targets.

[0012] Further, the first clustering module includes an association radius calculation unit, and the association radius calculation unit is used to calculate the association radius according to the speed of the research object.

[0013] Further, the first clustering module includes a density threshold calculation unit, which is used to calculate the density threshold according to the signal-to-noise ratio of the research object.

[0014] Further, the over-clustering judgment module includes a calculation unit and a comparison unit. The calculation unit is used to perform a point trace profile analysis on the points inside the clustering cluster and calculate the vehicle length and vehicle width information. The comparison unit is used to compare the vehicle width with a preset maximum vehicle width threshold. If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering. If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

[0015] Further, the first clustering module includes a clustering analysis unit. The clustering analysis unit compares the number of objects within the association radius neighborhood of the research object with the density threshold. If the number of objects is greater than the density threshold, the research object is a noise point. If the number of objects is less than or equal to the density threshold, the research object is a core point. Add the research object and the objects within the association neighborhood of the research object to the newly created cluster. Select a point from the newly created cluster. If the point is marked as unvisited, mark it as visited, and calculate the number of objects within the association radius neighborhood of this point. Compare the number of objects within the association radius neighborhood of this point with the density threshold. If the number of objects within the association radius neighborhood of this point is greater than the density threshold, this point is a core point, and add the objects within the neighborhood of this point to the newly created cluster. If the number of objects within the association radius neighborhood of this point is less than or equal to the density threshold, this point is a boundary point, and loop through until all points in the newly created cluster are marked as visited.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: A target tracking method based on DBSCAN and Kmeans provided by an embodiment of the present invention prevents the same target from being divided into multiple clusters by adding an over-clustering judgment to the clusters after DBSCAN clustering, improving the accuracy of the clustering result. And it solves the problem of obtaining the k parameter of the Kmeans algorithm through basic prior knowledge. This method is applicable to the target tracking problem when two or more vehicles are driving in parallel, can accurately perform target tracking, and greatly improves the accuracy of traffic flow statistics.

[0017] A target tracking system based on DBSCAN and Kmeans provided by an embodiment of the present invention prevents the same target from being divided into multiple clusters by adding an over-clustering judgment to the clusters after DBSCAN clustering, improving the accuracy of the clustering result. And it solves the problem of obtaining the k parameter of the Kmeans algorithm through basic prior knowledge. This system is applicable to the target tracking problem when two or more vehicles are driving in parallel, can accurately perform target tracking, and greatly improves the accuracy of traffic flow statistics. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings. In the drawings: Figure 1 It is a point track distribution diagram obtained by projecting point tracks onto a Cartesian coordinate system in a certain actual scenario; Figure 2 It is a flowchart of a target tracking method based on DBSCAN and Kmeans provided by the first embodiment of the present invention; Figure 3 It is a comparison diagram of the point cloud clustering effects of the target tracking method based on DBSCAN and Kmeans provided by the embodiments of the present invention and the existing method; Figure 4 It is a structural block diagram of a target tracking system based on DBSCAN and Kmeans provided by another embodiment of the present invention. Detailed implementation manners

[0019] To make the purpose, technical solutions and advantages of the present invention clearer, the following will further elaborate on the present invention in combination with the embodiments and drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and do not limit the present invention. Embodiment 1

[0020] As Figure 2 shown, a target tracking method based on DBSCAN and Kmeans provided by the embodiments of the present invention includes: Using a millimeter-wave radar to obtain a point cloud data set P of moving targets in a road scenario; Marking all point cloud data sets P as unvisited; Sorting all point cloud data sets P in descending order according to the signal-to-noise ratio; Sequentially selecting an unvisited object as the research object p and marking the research object p as visited; Setting an association radius Ɛ according to the speed of the research object p and setting a density threshold (MinPts) according to the signal-to-noise ratio of the research object p; Using the DBSCAN clustering method to perform clustering analysis on the research object and outputting the DBSCAN clustering cluster C; Using the prior target lateral dimension as a threshold to perform over-clustering judgment on the DBSCAN clustering cluster; If there is no over-clustering, output the DBSCAN clustering clusters; If there is over-clustering, calculate the possible number of targets; Use the Kmeans algorithm to perform secondary clustering on the DBSCAN clustering cluster C and output a new cluster of the number of targets.

[0021] Due to the conventional DBSCAN clustering algorithm, there is no priority difference for the objects to be clustered in the input. A random one is selected as a core point for the subsequent algorithm. This causes the situation that for some samples, the distances to two core objects may both be less than the association radius. However, these two core objects do not have density reachability and do not belong to the same clustering cluster, but the category of this sample needs to be defined. Generally speaking, the DBSCAN clustering algorithm follows the principle of first come, first served. The clustering category cluster that performs clustering first will mark this sample as its category. However, radar point cloud data has signal-to-noise ratio information. The higher the signal-to-noise ratio, the more accurate the measurement information (distance, speed, angle) of the point trace, and the more likely it is a target point. The lower the signal-to-noise ratio, the more likely it is a noise point. Therefore, starting from the points with high signal-to-noise ratio as core points for clustering operations can improve the clustering effect and ensure the stability of clustering. So, in this embodiment, first sort all point cloud data sets in descending order according to the signal-to-noise ratio, and then select a point in order as the research object.

[0022] In an actual traffic scenario, when the speed of a vehicle target is faster, the following distance between vehicles will be larger. Reflected in the point traces detected by the radar, it is that the distances between points of different targets are larger. Therefore, when clustering, the association radius Ɛ between point traces can be appropriately increased without worrying about clustering points of different targets into one cluster. On the contrary, when the vehicle speed is low, the distances between different target points detected by the radar are small, and the association radius Ɛ needs to be reduced to prevent clustering point traces of different targets into one cluster. Therefore, in this embodiment, the association radius is set according to the speed of the research object. The function of the association radius changing with speed is as follows: , where Ɛ is the association radius, with the unit of m, and v is the speed of the point trace, with the unit of m / s.

[0023] In the traffic application scenario, there are many types of targets to be tracked, such as cars, large vehicles, pedestrians, bicycles, etc. The point trace densities of different targets are inconsistent. Through the analysis of a large number of data, there is a simple understanding that the larger the size of the target, the higher the point trace density of the target, which also conforms to physical laws. At the same time, through the analysis of a large amount of data, generally, the RCS (radar cross section) of large-sized targets will also be larger, and reflected in the point traces, the signal-to-noise ratio will be greater. Therefore, the MinPts during clustering can be adjusted according to the signal-to-noise ratio. The MinPts value is set in 3 gears, namely gear 1, gear 3, and gear 5, as shown in the following formula:

[0024] , where MinPts is the density threshold, snr is the signal-to-noise ratio, and the unit is dB (decibel).

[0025] In this embodiment, the association radius is set according to the velocity of the traces, and the density threshold during clustering is adjusted according to the size of the signal-to-noise ratio, which can avoid clustering the traces of different targets into one cluster, and the output clustering cluster is the near-center position of the actual target, improving the accuracy of DBSCAN clustering.

[0026] In this embodiment, the specific method for clustering and analyzing the research object using the DBSCAN clustering method includes: Compare the number of objects N within the ε neighborhood of the research object p with the density threshold MinPts; If the number of objects N is greater than the density threshold MinPts, then the research object p is a noise point; If the number of objects N is less than or equal to the density threshold MinPts, then the research object p is a core point, and both the research object p and the objects within the ε neighborhood of the research object p are added to the newly created cluster c; Select a point p2 from the newly created cluster c. If the point is marked as unvisited, mark it as visited, and calculate the number of objects N2 within the ε neighborhood of p2, and compare the number of objects N2 with the density threshold MinPts; If N2 is greater than MinPts, then the point is a core point, and the objects within the neighborhood of the point are added to the newly created cluster c; If N2 is less than or equal to MinPts, then the point is a border point; Loop through until all points in the newly created cluster c are marked as visited; Output the DBSCAN clustering cluster C; Continue the above operations on the data set P until all points in the data set P are visited.

[0027] In this embodiment, the specific method for performing over-clustering judgment on the DBSCAN clustering cluster using the prior target lateral dimension as a threshold includes: Perform a trace profile analysis on the points inside the clustering cluster, and calculate the vehicle length and vehicle width information; Set the maximum vehicle width threshold, and compare the vehicle width with the maximum vehicle width threshold; If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering, the clustering cluster C is normal, and the clustering cluster C is directly output; If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

[0028] In the real world, the lengths of different vehicle models vary greatly, ranging from several meters to more than ten meters, while the width dimensions vary within a relatively small range, generally from 1.6m to 2.5m. This relatively small variation range can be used as prior knowledge to guide clustering. The specific steps are as follows: Perform point trace profile analysis on the points inside the DBSCAN clustering cluster C to calculate the vehicle length ysize and vehicle width xsize information. Set the maximum vehicle width index Thr = 2.5m. Compare xsize with Thr. If xsize ≤ Thr, it is considered that the vehicle width of the clustered cluster does not exceed the maximum limit, so it can be determined that the clustering is effective and multiple targets are not clustered together. If xsize > Thr, obviously the clustering result no longer meets the actual physical laws, and two or more targets must have been clustered together. Therefore, this clustering cluster C needs to be re - allocated into multiple clusters. Specifically, the number of targets clustered together can be determined by the integer part of xsize / Thr. That is, K = xsize / Thr. The number of targets calculated in this way is basically consistent with the actual situation through a large amount of data analysis. In this embodiment, by using the vehicle width to judge whether the cluster after DBSCAN clustering is over - clustered, it can effectively judge whether the target clustering is accurate and improve the accuracy of the clustering result.

[0029] Cluster the point traces in the clustering cluster C into k targets. Obviously, this is most easily achieved using the Kmeans algorithm, and since the value of k is known, the biggest defect of the Kmeans algorithm is avoided. Therefore, the subsequent process uses the Kmeans algorithm to re - allocate the clustering cluster C into K new clusters. The process of the Kmeans algorithm is as follows: Randomly select k points in the clustering cluster C as the initial mean centers , i = 1, 2…k; Calculate the distances from other sample points to each mean center. According to the principle of the closest distance, assign the sample points to k clusters and recalculate the mean center for each cluster , i = 1, 2…k; Judge whether the difference between mi and ui is less than the set threshold; If so, keep ui unchanged; If not, replace ui with mi; Judge whether all uis no longer change; If so, output as K clusters; If not, return and continue to execute the step of calculating the distances from other sample points to each point mean center.

[0030] Such as Figure 3As shown in the figure, a comparison chart of the point cloud clustering effect of the method of this embodiment and the existing DBSCAN algorithm is shown. It can be seen from the actual effect that the present invention can better adapt to the target clustering in the multi-vehicle parallel driving scenario compared with the existing method.

[0031] A target tracking method based on DBSCAN and Kmeans provided by an embodiment of the present invention improves the accuracy of the clustering result by adding over-clustering judgment to the clusters after DBSCAN clustering to prevent the same target from being split into multiple clusters. And the problem of obtaining the k parameter of the Kmeans algorithm is solved by basic prior knowledge. This method is applicable to the target tracking problem when two or more vehicles are driving in parallel, can accurately perform target tracking, and greatly improves the accuracy of traffic flow statistics. Embodiment 2

[0032] As Figure 4 As shown in the figure, a target tracking system based on DBSCAN and Kmeans provided by another embodiment of the present invention includes: a data acquisition module, a first clustering module, an over-clustering judgment module, and a second clustering module. The data acquisition module uses a millimeter-wave radar to acquire a point cloud data set of moving targets in a road scene; the first clustering module is used to mark all point cloud data sets as unvisited, sort all point cloud data sets in descending order according to the signal-to-noise ratio, select a point as the research object in order, mark the research object as visited, set an association radius according to the speed of the research object, set a density threshold according to the signal-to-noise ratio of the research object, and use the DBSCAN clustering method to perform clustering analysis on the research object, and output the DBSCAN clustering clusters; the over-clustering judgment module uses the prior target lateral dimension as a threshold to perform over-clustering judgment on the DBSCAN clustering clusters. If there is no over-clustering, the DBSCAN clustering clusters are output. If there is over-clustering, the possible number of targets is calculated; the second clustering module is used to perform secondary clustering on the DBSCAN clustering clusters using the Kmeans algorithm and output a new cluster of the number of targets.

[0033] Among them, the first clustering module includes an association radius calculation unit, and the association radius calculation unit is used to calculate the association radius according to the speed of the research object. The first clustering module includes a density threshold calculation unit, and the density threshold calculation unit is used to calculate the density threshold according to the signal-to-noise ratio of the research object.

[0034] The over-clustering judgment module includes a calculation unit and a comparison unit. The calculation unit is used to perform point trace profile analysis on the points inside the clustering cluster and calculate the vehicle length and vehicle width information; the comparison unit is used to compare the vehicle width with a preset maximum vehicle width threshold. If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering. If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

[0035] The first clustering module includes a clustering analysis unit. The clustering analysis unit compares the number of objects within the association radius neighborhood of the research object with the density threshold. If the number of objects is greater than the density threshold, the research object is a noise point. If the number of objects is less than or equal to the density threshold, the research object is a core point. Add the research object and the objects within the association neighborhood of the research object to the newly created cluster. Select a point from the newly created cluster. If the point is marked as unvisited, mark it as visited, and calculate the number of objects within the association radius neighborhood of this point. Compare the number of objects within the association radius neighborhood of this point with the density threshold. If the number of objects within the association radius neighborhood of this point is greater than the density threshold, this point is a core point, and add the objects within the neighborhood of this point to the newly created cluster. If the number of objects within the association radius neighborhood of this point is less than or equal to the density threshold, this point is a boundary point. Traverse in a loop until all points in the newly created cluster are marked as visited.

[0036] A target tracking system based on DBSCAN and Kmeans provided by an embodiment of the present invention prevents the same target from being divided into multiple clusters by adding over-clustering judgment to the clusters after DBSCAN clustering, improving the accuracy of the clustering result. And solves the problem of obtaining the k parameter of the Kmeans algorithm through basic prior knowledge. It is applicable to the target tracking problem when two or more vehicles are driving in parallel, can accurately perform target tracking, and greatly improves the accuracy of traffic flow statistics.

[0037] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A target tracking method based on DBSCAN and Kmeans, characterized in that, Including: Obtaining a point cloud data set of moving targets in a road scene using a millimeter-wave radar; Marking all point cloud data sets as unvisited; Sorting all point cloud data sets in descending order according to the signal-to-noise ratio; Selecting a point in order as the research object and marking the research object as visited; Setting an association radius according to the speed of the research object and setting a density threshold according to the signal-to-noise ratio of the research object; Performing clustering analysis on the research object using the DBSCAN clustering method and outputting DBSCAN clustering clusters; Using a prior target lateral dimension as a threshold to perform over-clustering judgment on the DBSCAN clustering clusters; If there is no over-clustering, output the DBSCAN clustering clusters; If there is over-clustering, calculate the possible number of targets; Performing secondary clustering on the DBSCAN clustering clusters using the Kmeans algorithm and outputting a new cluster of the number of targets.

2. The object tracking method based on DBSCAN and Kmeans according to claim 1, characterized in that, The specific method for setting the association radius according to the speed of the research object includes: The function of the association radius set according to the research object changing with speed is: ； where, Ɛ is the association radius, in m, and v is the speed of the trace, in m / s.

3. The target tracking method based on DBSCAN and Kmeans according to claim 2, wherein The specific method for setting the density threshold according to the signal-to-noise ratio of the research object includes: Setting the density threshold during clustering according to the signal-to-noise ratio, and the calculation formula for the value of the density threshold is: ； where, MinPts is the density threshold and snr is the signal-to-noise ratio, in dB.

4. The target tracking method based on DBSCAN and Kmeans according to claim 3, wherein The specific method for performing over-clustering judgment on the DBSCAN clustering clusters using a prior target lateral dimension as a threshold includes: Performing point trace profile analysis on the points inside the clustering cluster to calculate the vehicle length and vehicle width information; Setting a maximum vehicle width threshold and comparing the vehicle width with the maximum vehicle width threshold; If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering; If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

5. The target tracking method based on DBSCAN and Kmeans according to any one of claims 1-4, characterized in that, The specific method for performing clustering analysis on the research object using the DBSCAN clustering method includes: Comparing the number of objects within the association radius neighborhood of the research object with the density threshold; If the number of objects is greater than the density threshold, the research object is a noise point; If the number of objects is less than or equal to the density threshold, the research object is a core point, and the research object and the objects within its associated neighborhood are added to the newly created cluster; Selecting a point from the newly created cluster. If the point is marked as unvisited, mark it as visited, calculate the number of objects within the association radius neighborhood of this point, and compare the number of objects within the association radius neighborhood of this point with the density threshold; If the number of objects within the association radius neighborhood of this point is greater than the density threshold, this point is a core point, and the objects within the neighborhood of this point are added to the newly created cluster; If the number of objects within the association radius neighborhood of this point is less than or equal to the density threshold, this point is a boundary point; Perform a loop traversal until all points in the newly created cluster are marked as visited.

6. A target tracking system based on DBSCAN and Kmeans, characterized in that, Including: A data acquisition module, a first clustering module, an over-clustering judgment module, and a second clustering module, The data acquisition module obtains a point cloud data set of moving targets in a road scene using a millimeter-wave radar; The first clustering module is used to mark all point cloud data sets as unvisited, sort all point cloud data sets in descending order according to the signal-to-noise ratio, select a point in order as the research object, mark the research object as visited, set the association radius according to the speed of the research object, set the density threshold according to the signal-to-noise ratio of the research object, and use the DBSCAN clustering method to perform clustering analysis on the research object, and output the DBSCAN clustering clusters; The over-clustering judgment module uses the prior target lateral dimension as a threshold to judge the over-clustering of the DBSCAN clustering clusters. If there is no over-clustering, the DBSCAN clustering clusters are output. If there is over-clustering, the possible number of targets is calculated; The second clustering module is used to perform secondary clustering on the DBSCAN clustering clusters using the Kmeans algorithm and output new clusters of the number of targets.

7. The target tracking system based on DBSCAN and Kmeans according to claim 6, characterized in that, The first clustering module includes an association radius calculation unit, and the association radius calculation unit is used to calculate the association radius according to the speed of the research object.

8. The target tracking system based on DBSCAN and Kmeans according to claim 7, characterized in that, The first clustering module includes a density threshold calculation unit, and the density threshold calculation unit is used to calculate the density threshold according to the signal-to-noise ratio of the research object.

9. The target tracking system based on DBSCAN and Kmeans according to claim 8, characterized in that The over-clustering judgment module includes a calculation unit and a comparison unit. The calculation unit is used to perform point trace profile analysis on the points inside the clustering cluster and calculate the vehicle length and vehicle width information; The comparison unit is used to compare the vehicle width with the preset maximum vehicle width threshold. If the vehicle width information is less than or equal to the maximum vehicle width threshold, it is determined that there is no over-clustering. If the vehicle width information is greater than the maximum vehicle width threshold, it is determined that there is over-clustering.

10. The object tracking system based on DBSCAN and Kmeans according to any one of claims 6-9, characterized in that, The first clustering module includes a clustering analysis unit. The clustering analysis unit compares the number of objects in the neighborhood of the association radius of the research object with the density threshold. If the number of objects is greater than the density threshold, the research object is a noise point. If the number of objects is less than or equal to the density threshold, the research object is a core point. Add the research object and the objects in the association neighborhood of the research object to the newly created cluster. Select a point from the newly created cluster. If the point is marked as unvisited, mark it as visited, and calculate the number of objects in the neighborhood of the association radius of the point. Compare the number of objects in the neighborhood of the association radius of the point with the density threshold. If the number of objects in the neighborhood of the association radius of the point is greater than the density threshold, the point is a core point, and add the objects in the neighborhood of the point to the newly created cluster. If the number of objects in the neighborhood of the association radius of the point is less than or equal to the density threshold, the point is a boundary point, and loop through until all points in the newly created cluster are marked as visited.

Images