Radar point cloud clustering method and system based on vehicle size adaptive discrimination

Abstract

The invention discloses a radar point cloud clustering method and system based on vehicle model size adaptive discrimination. The method comprises the following steps: setting parameters of each level of vehicle model size range during clustering, and a corresponding clustering point cloud number threshold value and a clustering speed difference threshold value; obtaining radar point cloud data and sorting to obtain a point cloud data set; traversing reflection points in the point cloud data set in sequence to perform clustering based on the reflection points, and performing point cloud clustering with the vehicle size range expanded step by step by taking the current reflection point as a center to obtain a clustering cluster with the current reflection point as a clustering core; screening out the point cloud data in the clustering cluster from the point cloud data set, and continuing to perform clustering based on the reflection points; and after traversing the point cloud data set, obtaining a plurality of clusters and corresponding vehicle model sizes. According to the method, point cloud clustering with the vehicle type size range expanded step by step is carried out according to the vehicle type size increasing sequence, the point cloud of the large vehicle and the point cloud of the small vehicle can be effectively and accurately clustered, and the problem of splitting of large vehicle clustering is avoided.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a radar point cloud clustering method and system based on vehicle size adaptive discrimination. Background Art

[0002] Traffic millimeter-wave radar can detect road vehicles and obtain radar point clouds after signal processing. Radar point clouds are cloud-like point clouds that reflect electromagnetic waves from different parts of the vehicle. For vehicles of different sizes, the point clouds obtained by the radar are also of different sizes, and when the traffic is dense, the point clouds of vehicles of different sizes are also close together. The original radar point cloud needs to be clustered to identify different vehicles from a group of similar point clouds. The most common radar clustering algorithm is DBSCAN (Density-Based Spatial Clustering of Applications with Noise), which is a density-based clustering algorithm. However, in scenes where large and small vehicles are mixed, the DBSCAN algorithm often has difficulty effectively distinguishing the point clouds of large vehicles from those of small vehicles, which makes it easy to cluster the point clouds of large vehicles into two classes, that is, the problem of large vehicle splitting occurs.

[0003] The "A millimeter wave radar point cloud clustering method and system" disclosed in the Chinese patent literature has a publication number of CN118549888A and a publication date of 2024.8.27, including: obtaining point cloud data containing speed, azimuth, and pitch angle, and performing multi-frame prediction and superposition on the point cloud data to obtain superimposed data; filtering the superimposed data by signal-to-noise ratio parameters to obtain filtered data; performing DBSCAN clustering on the filtered data to obtain the number of data clusters, and using the number of data clusters as the number of human targets. This technology provides a millimeter wave radar point cloud clustering method and system, which optimizes the accuracy of radar target point cloud clustering and the accuracy of the number of indoor human targets by superimposing n frames of point cloud speed prediction and filtering by signal-to-noise ratio parameters, and reduces the recognition rate of false targets. However, this technology is aimed at clustering and identifying indoor human targets. There are no mixed scenes with huge size differences such as large vehicles and small vehicles. Therefore, there is no premise for the split when clustering large vehicles, and the corresponding problem cannot be solved. In addition, this technology also performs clustering through the DBSCAN clustering algorithm, which also makes it difficult to effectively distinguish the point cloud of large vehicles from that of small vehicles, resulting in the point cloud of large vehicles being easily clustered into two classes, thus causing the problem of large vehicle splitting. Summary of the invention

[0004] The present invention aims to overcome the problem that when a common radar clustering algorithm is used in the prior art to cluster the millimeter-wave radar point clouds of vehicles, it is difficult to effectively distinguish the point clouds of large vehicles from those of small vehicles, and the point clouds of large vehicles are easily clustered into two classes, resulting in the splitting of large vehicles. A radar point cloud clustering method and system based on adaptive discrimination of vehicle type size are provided.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A radar point cloud clustering method based on vehicle size adaptive discrimination comprises: Set the parameters of the size range of each level of vehicle models during clustering, as well as the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; obtain radar point cloud data and sort them to obtain a point cloud data set; The reflection points in the point cloud data set are traversed in order to perform clustering based on the reflection points, and the point cloud clustering with the vehicle size range gradually expanded with the current reflection point as the center is performed to obtain a cluster cluster with the current reflection point as the cluster core; Screening out the point cloud data in the cluster from the point cloud data set, and continuing to perform clustering based on reflection points; After traversing the point cloud dataset, several clusters and corresponding vehicle model sizes are obtained.

[0006] The radar point cloud clustering method based on adaptive vehicle size discrimination in the present invention can perform clustering by gradually expanding the selection range from small to large vehicle types according to different actual vehicle types. First, clustering is performed within the range of small-sized vehicle types. If the clustering conditions are met, clustering is completed. If the clustering conditions are not met, the range of vehicle type sizes is further expanded for clustering. The clustering is performed step by step until the range of the maximum vehicle type size is reached. Since the prerequisite of the vehicle type size is already met in the clustering process, radar point cloud clustering can be performed accurately according to different vehicle type sizes. The accuracy of the clustering is determined by the clustering point cloud quantity threshold and the clustering speed difference threshold in the clustering conditions, and the large vehicle splitting problem similar to the existing density clustering method will not occur.

[0007] Preferably, the acquiring radar point cloud data and sorting to obtain a point cloud data set comprises: The signal strength of the reflection points in the point cloud data is ranked from large to small as the first priority; The density values ​​of the reflection points are ranked from large to small as the second priority.

[0008] Preferably, the clustering based on reflection points includes: setting the size range of vehicles from level 1 to level N from small to large; when performing point cloud clustering of the size range of vehicles of level i with the current reflection point as the center, screening the number of point clouds Mi of reflection points that meet the size judgment standard of vehicles of level i; If Mi is greater than the clustering point cloud quantity threshold of the i-level vehicle size range, point cloud clustering of the next level vehicle size range is performed; otherwise, a cluster cluster with the current reflection point as the clustering core and the corresponding i-level vehicle size is obtained.

[0009] Preferably, the point cloud clustering of the i-level vehicle size range with the current reflection point as the center includes: Taking the coordinates of the current reflection point as the center and the speed direction of the current reflection point as the length direction, a rectangular frame with a length of Li and a width of Wi is constructed to obtain the size range of the i-level vehicle model at the current reflection point; Point cloud clustering is performed within the i-class vehicle size range of the current reflection point.

[0010] Preferably, the i-level vehicle size judgment criteria include: The reflection point is within the i-class vehicle size range of the current reflection point; and The absolute value of the speed difference between the reflection point and the current reflection point is less than the cluster speed difference threshold of the i-class vehicle size range; The speed difference is a difference in projection speed in a direction in which the road extends.

[0011] Preferably, the parameters of the size range of each level of vehicle models when setting clustering include: Obtain historical radar point cloud data and filter out point cloud data of each level of vehicle size; for level i vehicle size, obtain the size of all point clouds of level i vehicle size and the number of reflection points contained in each point cloud; Arrange the sizes of all point clouds in descending order, and select the size at the size setting percentile as the parameter of the size range of the i-level vehicle model.

[0012] Preferably, the number of reflection points contained in each point cloud of the i-level vehicle size is arranged in descending order, and the number of reflection points at the set quantity quantile is selected as the cluster point cloud quantity threshold of the i-level vehicle size range; The maximum speed difference between the reflection points in each point cloud based on the i-level vehicle size is sorted in descending order, and the maximum speed difference at the speed setting percentile is selected as the clustering speed difference threshold of the i-level vehicle size range; The speed difference is a difference in projection speed in a direction in which the road extends.

[0013] Preferably, the clustering based on reflection points further includes initial screening of point clouds; Set the point cloud initial screening range centered on the current reflection point, where the point cloud initial screening range includes the maximum range of the size range of each level of vehicle models; Filter out a set of reflection points within the initial screening range of the point cloud from the radar point cloud data; Point cloud clustering is performed based on the reflection point set with a gradually expanding vehicle size range.

[0014] As a preference, noise clustering judgment is performed on the clustering clusters that have completed clustering: When the number of reflection points in a cluster is less than the noise point threshold, it is judged as a noise cluster; After traversing the point cloud data set, the clusters judged as noise clusters are discarded, and several clusters and corresponding vehicle model sizes are obtained.

[0015] A radar point cloud clustering system based on vehicle size adaptive discrimination comprises: The historical point cloud saving module saves historical radar point cloud data, sets the parameters of the size range of each level of vehicle models, and sets the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; A radar point cloud acquisition module is used to acquire radar point cloud data to be clustered; The step-by-step clustering module traverses the reflection points in the point cloud dataset to perform clustering based on the reflection points, and performs point cloud clustering with a gradually expanding vehicle size range with the current reflection point as the center.

[0016] The radar point cloud clustering system in the present invention can be combined as a subsystem or a large module into the entire traffic management system or other main systems to provide the final vehicle point cloud clustering results; the historical point cloud module saves historical radar point cloud data and needs to manually determine the point cloud range corresponding to different vehicle sizes in advance, and uses it as a preliminary sample to determine the parameters of the vehicle size range and the corresponding clustering judgment conditions. At the same time, the radar point cloud data can be continuously updated to obtain more accurate parameters and clustering judgment conditions, thereby improving the accuracy of the final vehicle clustering; the radar point cloud acquisition module can acquire the radar point clouds of vehicles in the corresponding area in real time, thereby performing clustering; the step-by-step clustering module is the part of the present invention that specifically executes the point cloud clustering method, and together with the parameters and clustering judgment conditions set by the historical point cloud storage module, it realizes the point cloud clustering of the present invention that accurately distinguishes between large and small vehicles, and avoids the splitting of large vehicles during the clustering process.

[0017] The present invention has the following beneficial effects: determining the size ranges of different vehicle models and corresponding clustering judgment conditions based on historical radar point cloud data; performing point cloud clustering with a step-by-step expansion of the vehicle model size range in the order of increasing vehicle model size, and clustering reflection points based on the clustering judgment conditions of the vehicle model size ranges at each level, which can effectively and accurately cluster the point clouds of large vehicles and small vehicles, and avoid the problem of splitting the clustering of large vehicles; when performing clustering based on reflection points, first perform a preliminary screening of the point cloud with the current reflection point as the center, and perform subsequent step-by-step clustering processes with the range after the preliminary screening, which can effectively reduce the amount of calculation in the clustering process; sorting the radar point cloud data according to the set priority rules, the front reflection point can better reflect the radar reflection point of the vehicle, and clustering according to the priority order can obtain a more accurate vehicle clustering result. Brief Description of the Drawings

[0018] Figure 1 is a flowchart of the radar point cloud clustering method based on adaptive discrimination of vehicle model size in the present invention.

[0019] Figure 2 is a flowchart of the point cloud clustering for gradually expanding the vehicle model size range in the present invention.

[0020] Figure 3 is a schematic diagram of the radar point cloud clustering system based on adaptive discrimination of vehicle model size in the present invention.

[0021] Figure 4 is a schematic diagram of the vehicle model size range centered on the current reflection point in the present invention.

[0022] In the figure: 1. Historical point cloud storage module; 2. Radar point cloud acquisition module; 3. Gradual clustering module; 4. Clustering result output module; 30. Sorting unit; 31. First-level clustering unit; 32. Second-level clustering unit; 33. N-level clustering unit. Detailed Embodiment

[0023] The following further describes the present invention in conjunction with the drawings and the detailed embodiment.

[0024] The original radar point cloud needs to be clustered to identify each different vehicle from clusters of similar points. The most common current radar clustering algorithm is DBSCAN (Density-Based Spatial Clustering of Applications with Noise), which is a density-based clustering algorithm. It defines a cluster as the largest set of density-connected points, can divide regions with sufficient high density into clusters, and can discover clusters of any shape in a spatial database with noise. DBSCAN requires two parameters: scanning radius (eps) and minimum number of included points (minPts). Starting from an unvisited point arbitrarily, find all nearby points whose distance from it is within eps (including eps). If the number of nearby points ≥ minPts, the current point and its nearby points form a cluster, and the starting point is marked as visited. Then recursively, process all unmarked points in the cluster in the same way to expand the cluster. If the number of nearby points < minPts, the point is temporarily marked as a noise point. If the cluster is fully expanded, that is, all points in the cluster are marked as visited, then use the same algorithm to process unvisited points.

[0025] Density clustering is clustering performed after determining the scanning radius and the minimum number of included points. In the specific vehicle radar point cloud scanning process, the point cloud of a small car is relatively concentrated. By setting the scanning radius and the minimum number of included points in the density-based clustering algorithm, the point cloud of the small car can be clustered in one cluster more reasonably. However, for a large car, due to its longer length, the point cloud of a large car will be more concentrated in the front and rear parts in the actual point cloud. With the same scanning radius and the minimum number of included points as parameter settings, it is easy to cluster the rear point cloud and the front point cloud of the large car into two clusters, thereby causing the problem of large car splitting. Based on the above problems, the present invention provides an improved vehicle point cloud clustering method to overcome the problem of large car splitting during clustering.

[0026] like Figure 1 As shown, a radar point cloud clustering method based on vehicle size adaptive discrimination includes: Set the parameters of the size range of each level of vehicle models during clustering, as well as the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; obtain radar point cloud data and sort them to obtain a point cloud data set; The reflection points in the point cloud data set are traversed in order to perform clustering based on the reflection points, and the point cloud clustering with the vehicle size range gradually expanded with the current reflection point as the center is performed to obtain a cluster cluster with the current reflection point as the cluster core; Screening out the point cloud data in the cluster from the point cloud data set, and continuing to perform clustering based on reflection points; After traversing the point cloud dataset, several clusters and corresponding vehicle model sizes are obtained.

[0027] It should be noted that the radar point cloud clustering method based on adaptive vehicle size discrimination in the present invention can expand the selection range from small to large according to the actual vehicle models, and cluster them first with the range of small-sized vehicles. If the clustering conditions are met, clustering is completed. If the clustering conditions are not met, the range of vehicle sizes is further expanded for clustering; the expansion is performed step by step until the range of the maximum vehicle size is reached to complete the clustering; since the prerequisite of vehicle size is already met in the clustering process, radar point clouds can be accurately clustered according to different vehicle sizes, and the accuracy of the clustering is determined by the clustering point cloud quantity threshold and the clustering speed difference threshold in the clustering conditions, and the large vehicle splitting problem similar to the existing density clustering and the like will not occur.

[0028] Specifically, the parameters of the size range of each level of vehicle models during clustering, as well as the corresponding cluster point cloud quantity threshold and cluster speed difference threshold are pre-set before clustering, that is, each level of vehicle model size range corresponds to its cluster point cloud quantity threshold and cluster speed difference threshold, which together constitute the clustering judgment condition. The clustering judgment condition is used to determine whether the clustering requirements of the current level of vehicle model size range are met. The specific settings can be determined by expert experience or obtained by analyzing historical data.

[0029] When clustering begins, the radar point cloud data to be clustered is obtained first, and the reflection points in the radar point cloud are sorted according to the preset priority order to obtain the point cloud data set, so as to determine the order priority of each reflection point as the cluster core point during clustering. The reflection points are selected in order for clustering based on the reflection points.

[0030] When clustering, the current reflection point is used as the center. First, the reflection points are screened from the smallest vehicle size range. If the screened reflection points meet the clustering conditions, clustering is completed, and the screened reflection points and the current reflection points form a cluster cluster; if the clustering conditions are not met, the reflection points are screened from the next level of vehicle size range, and the vehicle size range is expanded step by step for clustering until the screened reflection points meet the clustering conditions and the cluster position is completed. If the maximum vehicle size range has been reached for screening, the screened reflection points and the current reflection points are directly used to form a cluster cluster, and the clustering conditions are no longer judged, because the minimum vehicle size and the maximum vehicle size are determined according to the actual vehicle size. Generally, there is no vehicle that exceeds the maximum vehicle size, so there is no need to increase the vehicle size range for clustering.

[0031] After completing the clustering of a cluster, all reflection points contained in the cluster are screened out from the point cloud data set to form a new point cloud data set and the corresponding sorting, and then continue to traverse the reflection points, starting from the reflection point at the front, and clustering in the same way until the point cloud data set is empty. For several clusters obtained by clustering, the corresponding cluster results and the corresponding vehicle model size are output.

[0032] As an optional embodiment, noise clustering judgment is performed on the clustering clusters that have completed clustering: When the number of reflection points in a cluster is less than the noise point threshold, it is judged as a noise cluster; After traversing the point cloud data set, the clusters judged as noise clusters are discarded, and several clusters and corresponding vehicle model sizes are obtained.

[0033] It should be noted that the collected radar point cloud data will inevitably be interfered by noise, that is, there is noise in the cluster clusters that are finally clustered. Therefore, after each clustering, it is necessary to perform noise clustering judgment on the cluster clusters, eliminate the noise clusters, and obtain the cluster clusters corresponding to the real different vehicle sizes and the corresponding vehicle sizes. In the present invention, the judgment of whether it is a noise cluster is also determined by the number of reflection points in the cluster cluster. If the number of reflection points is large, it means that there are a large number of reflection points in the same state, so it is very likely to be a point cloud of a vehicle; if the number of reflection points is very small, it means that the reflection point states are different, and it is very likely to be noise interference. Therefore, the noise point threshold can be determined based on actual data or historical experience. When the number of reflection points is less than the noise point threshold, it is judged as a noise cluster.

[0034] Specifically, when clustering, the current reflection point is used as the center, and the reflection points are first screened from the smallest vehicle size range. If the screened reflection points meet the clustering conditions, clustering is completed, and the screened reflection points and the current reflection points are combined into a cluster cluster; if the clustering conditions are not met, the reflection points are screened from the next vehicle size range, and the vehicle size range is expanded step by step for clustering until the screened reflection points meet the clustering conditions and the cluster position is completed. If the maximum vehicle size range has been reached for screening, the screened reflection points and the current reflection points are directly combined into a cluster cluster, and the clustering conditions are no longer judged, because the minimum vehicle size and the maximum vehicle size are determined according to the actual vehicle size. Generally, there is no vehicle that exceeds the maximum vehicle size, so there is no need to increase the vehicle size range for clustering.

[0035] After completing the clustering of a cluster, determine the total number of reflection points in the cluster. If the total number of reflection points is less than the noise point threshold, it means that this cluster is a noise cluster and the noise cluster is identified. If the total number of reflection points is greater than the noise point threshold, it is considered to be a non-noise cluster, confirming that the cluster is the clustering result of the vehicle and correspondingly identifying the vehicle model size.

[0036] After the noise clustering is completed, all reflection points contained in the cluster are screened out from the point cloud data set to form a new point cloud data set and the corresponding sorting. Then, the reflection points are traversed and clustering is performed in the same way starting from the reflection point at the front until the point cloud data set is empty. For the clusters obtained by clustering, the clusters marked as noise clusters are discarded, and the clustering results of the remaining non-noise clusters and the corresponding vehicle model sizes are output.

[0037] As a specific embodiment, acquiring radar point cloud data and sorting to obtain a point cloud data set includes: The signal strength of the reflection points in the point cloud data is ranked from large to small as the first priority; The density values ​​of the reflection points are ranked from large to small as the second priority.

[0038] It should be noted that the radar point cloud data is sorted according to the set priority rules. The front reflection point is more likely to be the radar reflection point of the vehicle, and the back reflection point is more likely to be noise. Clustering according to the priority order can obtain more accurate vehicle clustering results. For the selection of priority, the signal strength of the reflection point and the density of the reflection point are the two optimal options. Other parameters related to the reflection point can also be used as the basis for priority sorting.

[0039] Specifically, obtain the radar point cloud data composed of a large number of reflection points collected by the current radar, where each reflection point has a signal strength RCS value. The larger the RCS value, the stronger the signal of the reflection point. In addition, for each reflection point, calculate the density value of the point. The density value is defined as the number of reflection points contained in a circle with the reflection point as the center and a set distance (in meters) as the radius. Then, all reflection points in the radar point cloud data are sorted based on the principle of signal strength value from large to small and then density value from large to small. This sorting is to obtain the order priority of each reflection point when it is used as the core point of clustering during clustering, and obtain the sorted point cloud data set.

[0040] As a specific embodiment, clustering based on reflection points includes: Set the size range of vehicles from Class 1 to Class N from small to large; When clustering the point cloud of the i-level vehicle size range with the current reflection point as the center, the number of point clouds Mi of the reflection points that meet the i-level vehicle size judgment standard is selected; If Mi is greater than the clustering point cloud quantity threshold of the i-level vehicle size range, point cloud clustering of the next level vehicle size range is performed; otherwise, a cluster cluster with the current reflection point as the clustering core and the corresponding i-level vehicle size is obtained.

[0041] It should be noted that the clustering based on reflection points in the present invention is to cluster radar point clouds by gradually expanding the vehicle size range after determining the vehicle size range with the reflection point as the geometric center; the classification of vehicle size ranges can be divided according to the actual vehicle types, with the 1st level vehicle size range corresponding to the smallest vehicle size, the 2nd level vehicle size range corresponding to the next level vehicle size, and the Nth level vehicle size range corresponding to the largest vehicle size. When the number of point clouds that meet the vehicle size judgment standard in the previous level vehicle size range does not meet the requirements, it means that it is not suitable for clustering in the previous level vehicle size range, and clustering of the next level vehicle size range is required until the requirements are met.

[0042] It is worth noting that since the maximum vehicle size range is N-level vehicle size range, when clustering is performed in the vehicle size range from level 1 to level N-1, if the number of point clouds that meet the corresponding vehicle size judgment standard does not meet the requirements, clustering of the next level vehicle size range can be performed; however, when clustering is performed in the N-level vehicle size range based on the current reflection point, since N is already the maximum value, there is no need to determine whether the next level of expansion is needed based on the number of clustering points. Instead, the screened reflection points that meet the conditions are directly used to form a cluster cluster with the current reflection point as the clustering core.

[0043] Specifically, Figure 2 Shown is a flowchart for point cloud clustering with a gradually expanding vehicle size range.

[0044] Before clustering, the parameters of the size range of each level of vehicle models during clustering, as well as the corresponding cluster point cloud quantity threshold and cluster speed difference threshold are pre-set. That is, each level of vehicle model size range corresponds to its cluster point cloud quantity threshold and cluster speed difference threshold, which together constitute the clustering judgment condition.

[0045] When clustering begins, the radar point cloud data to be clustered is obtained first, and the reflection points in the radar point cloud are sorted according to the preset priority order to obtain the point cloud data set, so as to determine the order priority of each reflection point as the cluster core point during clustering. The reflection points are selected in order for clustering based on the reflection points.

[0046] The point cloud of the level 1 vehicle size range is clustered with the current reflection point as the center, and the point cloud number M1 of the reflection point that meets the level 1 vehicle size judgment standard is screened; then the number of point clouds that meet the requirements M1 is compared with the threshold of the number of clustered point clouds of the level 1 vehicle size range. When M1 is less than or equal to the threshold of the number of clustered point clouds of the level 1 vehicle size range, the clustering process is completed, and a cluster cluster with the current reflection point as the cluster core and containing M1 reflection points that meet the conditions and the corresponding vehicle size (level 1 vehicle size range) is obtained. When M1 is greater than the threshold of the number of clustered point clouds of the level 1 vehicle size range, it is expanded and then clustered.

[0047] The point cloud of the level 2 vehicle size range is clustered with the current reflection point as the center, and the point cloud number M2 of the reflection point that meets the level 2 vehicle size judgment standard is screened; then the number of point clouds that meet the requirements M2 is compared with the threshold of the number of clustered point clouds of the level 2 vehicle size range. When M2 is less than or equal to the threshold of the number of clustered point clouds of the level 2 vehicle size range, the clustering process is completed, and a cluster cluster with the current reflection point as the cluster core and containing M2 reflection points that meet the conditions and the corresponding vehicle size (level 2 vehicle size range) is obtained. When M2 is greater than the threshold of the number of clustered point clouds of the level 2 vehicle size range, it is expanded and then clustered.

[0048] The same process is followed to expand clustering step by step until the point cloud of N-level vehicle size range is clustered with the current reflection point as the center, and the point cloud number MN of reflection points that meet the N-level vehicle size judgment standard is screened out, and a cluster cluster with the current reflection point as the clustering core and containing MN satisfying reflection points and the corresponding vehicle size (N-level vehicle size range) is obtained.

[0049] Furthermore, the point cloud clustering of the i-level vehicle size range with the current reflection point as the center includes: Taking the coordinates of the current reflection point as the center and the speed direction of the current reflection point as the length direction, a rectangular frame with a length of Li and a width of Wi is constructed to obtain the size range of the i-level vehicle model at the current reflection point; Point cloud clustering is performed within the i-class vehicle size range of the current reflection point.

[0050] It should be noted that for the determination of the i-level vehicle size range centered on the current reflection point, the shape of the vehicle size range can be different shapes such as circle, square or matrix. Clustering based on reflection points is performed on the basis of determining the specific vehicle size range. However, different vehicle size range selections have a certain impact on the final clustering accuracy. Since the horizontal cross-section of the vehicle is generally rectangular, choosing a rectangle as the vehicle size range can better fit the actual vehicle size and avoid missing reflection points or mixing noise during the clustering process.

[0051] Specifically, Figure 4 The figure shows a schematic diagram of the vehicle size range centered on the current reflection point. The coordinates are established with the extension direction of the road as the X-axis direction and the road width direction as the Y-axis direction. The actual speed of the reflection point P can be decomposed into the superposition of the speed of the X-axis projection and the speed of the Y-axis projection, thereby obtaining the speed and direction A of the reflection point P.

[0052] Next, a rectangular frame with a length of L and a width of W is constructed with the coordinates of the reflection point P as the center and the speed direction A of the reflection point P as the direction. It can be considered that the rectangular frame actually represents the size of the vehicle body of the reflection point P which is currently the core point of the clustering, and subsequent clustering will also be performed within the rectangular frame.

[0053] Optional, i-class vehicle size criteria include: The reflection point is within the i-class vehicle size range of the current reflection point; and The absolute value of the speed difference between the reflection point and the current reflection point is less than the cluster speed difference threshold of the i-class vehicle size range; The speed difference is the difference in projected speed in the direction in which the road extends.

[0054] It should be noted that the specific vehicle size judgment standard is determined in accordance with the size range of different vehicle types, that is, each vehicle type size range has its own vehicle size judgment standard and its own clustering point cloud quantity threshold. The i-level vehicle size judgment standard can be regarded as a way to cluster the point cloud within the i-level vehicle size range centered on the current reflection point, thereby replacing the existing density-based point cloud clustering.

[0055] It is worth noting that for the i-class vehicle size judgment standard, the first step is to screen all reflection points within the i-class vehicle size range centered on the current reflection point; after the screening is completed, the speed difference between the screened reflection points and the current reflection point is judged, and all reflection points whose absolute value of the speed difference is less than the clustering speed difference threshold of the i-class vehicle size range are selected as the final result.

[0056] Since the speed of the reflection point is actually the superposition of the speed of the X-axis and the Y-axis, the direction of the actual speed of different reflection points may have a certain deviation. In order to facilitate comparison and calculation, based on the fact that vehicles basically travel along the extension direction of the road, the projection speed of the speed of the reflection point in the X-axis direction is selected as the parameter reflecting the vehicle speed, and the difference in the projection speed in the X-axis direction between the two reflection points is used as the speed difference between the two reflection points. When the absolute value of the speed difference is less than the clustering speed difference threshold of the i-level vehicle size range, it means that the speeds between the two reflection points are very close, which meets the judgment range of belonging to the same vehicle and can be clustered.

[0057] As a specific embodiment, the parameters for the size range of each level of vehicle model when setting clustering include: Obtain historical radar point cloud data and filter out point cloud data of each level of vehicle size; for level i vehicle size, obtain the size of all point clouds of level i vehicle size and the number of reflection points contained in each point cloud; Arrange the sizes of all point clouds in descending order, and select the size at the size setting percentile as the parameter of the size range of the i-level vehicle model.

[0058] It should be noted that the determination of the parameters for the vehicle size range during clustering, as well as the determination of the clustering point cloud quantity threshold and the clustering speed difference threshold for a certain vehicle size range, are important factors affecting the accuracy of the final clustering results of the vehicle; therefore, in order to be more in line with the actual data of the vehicle, the historical radar point cloud data of the vehicle can be used to analyze the vehicle size range and the corresponding clustering point cloud quantity threshold and clustering speed difference threshold.

[0059] Furthermore, the number of reflection points contained in each point cloud based on the i-level vehicle size is arranged in descending order, and the number of reflection points at the set number quantile is selected as the cluster point cloud number threshold for the i-level vehicle size range; The maximum speed difference between the reflection points in each point cloud based on the i-level vehicle size is sorted in descending order, and the maximum speed difference at the speed setting percentile is selected as the clustering speed difference threshold of the i-level vehicle size range; The speed difference is the difference in projected speed in the direction in which the road extends.

[0060] It is worth noting that the historical radar point cloud data can be obtained through pre-processing, that is, first determining the vehicle sizes of different levels, collecting radar point clouds for each level of vehicle size, thereby obtaining a large amount of point cloud data corresponding to the vehicle sizes of different levels, and analyzing the large amount of point cloud data for each level of vehicle size.

[0061] Specifically, the length and width of the rectangular frame corresponding to the vehicle size range and the clustering parameters related to the rectangular frame are calibrated based on the following method: Obtain multiple batches of radar point cloud data from traffic millimeter-wave radars in various scenarios, and filter out point cloud data of different levels of vehicle sizes from the point cloud data. Count the size of the point cloud of each level of vehicle size and the number of reflection points contained in the point cloud. For the size of the point cloud, the minimum enclosing rectangle surrounding all reflection points can be constructed with the reflection points in the point cloud, and the length and width of the minimum enclosing rectangle are used as the length and width of the vehicle size.

[0062] For a certain level of vehicle size, the size data of its point cloud are arranged in ascending order, and then the size data at the 0.9 percentile is taken as the size data of the clustering rectangular box of the vehicle size of this level; the size data of the clustering rectangular box of other levels of vehicle size are determined in the same way.

[0063] For a certain level of vehicle size, the number of reflection points contained in its point cloud is arranged in ascending order, and then the number of reflection points at the 0.8 percentile is taken as the cluster point cloud number threshold for this level of vehicle size; the cluster point cloud number threshold for other levels of vehicle size is determined in the same way.

[0064] For a certain level of vehicle size, the maximum speed difference between the X-axis speed values ​​of each reflection point in its each point cloud data is calculated respectively, and each speed difference maximum value is arranged in ascending order, and then the speed difference value at the 0.9 percentile is taken as the clustering speed difference threshold of the vehicle size of this level; the clustering speed difference threshold of other levels of vehicle size is determined in the same way.

[0065] As an optional embodiment, clustering based on reflection points also includes initial screening of point clouds; Set the initial screening range of the point cloud centered on the current reflection point. The initial screening range of the point cloud includes the maximum range of the size range of each level of vehicle models; filter out the set of reflection points within the initial screening range of the point cloud from the radar point cloud data; Point cloud clustering is performed based on the reflection point set with a gradually expanding vehicle size range.

[0066] It should be noted that when clustering the point cloud with the current reflection point as the center and the vehicle size range is gradually expanded, for each level of vehicle size range, it is necessary to find suitable clustering points from the entire radar point cloud data, which will greatly increase the calculation amount of the entire clustering process. In order to reduce the actual calculation amount, the entire radar point cloud data can be preliminarily screened based on the reflection point. Since the point cloud that meets the clustering requirements in the actual clustering process is the reflection point near the current reflection point, the point cloud preliminarily screening range centered on the current reflection point can be determined first to obtain the reflection point set after the preliminarily screening. The subsequent clustering process is based on the reflection points in the reflection point set, which can greatly reduce the calculation amount of the clustering process.

[0067] It is worth mentioning that in order to avoid the vehicle size range exceeding the point cloud initial screening range during the clustering process of the gradually expanding vehicle size range, the actual point cloud initial screening range centered on the current reflection point is larger than the maximum range of the vehicle size range at each level. Therefore, no matter which level of vehicle size range is used for clustering, no reflection point will be missed.

[0068] As a specific embodiment, the present invention also provides Figure 3 A radar point cloud clustering system based on vehicle size adaptive discrimination is shown, comprising: The historical point cloud saving module 1 saves the historical radar point cloud data, sets the parameters of the size range of each level of vehicle models, and sets the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; Radar point cloud acquisition module 2, acquiring radar point cloud data to be clustered; The step-by-step clustering module 3 traverses the reflection points in the point cloud data set to perform clustering based on the reflection points, and performs point cloud clustering with a gradually expanding vehicle size range with the current reflection point as the center.

[0069] It should be noted that the radar point cloud clustering system in the present invention can be combined as a subsystem or a large module into the entire traffic management system or other main system to provide the final vehicle point cloud clustering results; the historical point cloud module saves historical radar point cloud data and needs to manually determine the point cloud range corresponding to different vehicle sizes in advance, and use it as a preliminary sample to determine the parameters of the vehicle size range and the corresponding clustering judgment conditions. At the same time, the radar point cloud data can be continuously updated to obtain more accurate parameters and clustering judgment conditions, thereby improving the accuracy of the final vehicle clustering; the radar point cloud acquisition module can acquire the radar point cloud of the vehicles in the corresponding area in real time, so as to perform clustering; the step-by-step clustering module is the part of the present invention that specifically executes the point cloud clustering method, and it and the parameters and clustering judgment conditions set by the historical point cloud storage module together realize the point cloud clustering that accurately distinguishes between large and small vehicles in the present invention, and avoids the splitting of large vehicles during the clustering process.

[0070] Specifically, the radar point cloud obtained in the radar point cloud acquisition module 2 can be saved in the historical point cloud storage module 1 after manual verification and comparison and annotation, and saved as historical radar point cloud data to assist in adjusting the parameters of the size range of each level of vehicle model and the corresponding clustering point cloud quantity threshold and clustering speed difference threshold. At the same time, the radar point cloud data in the radar point cloud acquisition module 2 is input into the step-by-step clustering module 3 to perform point cloud clustering of the present invention with the current reflection point as the center and the vehicle model size range gradually expanded.

[0071] The step-by-step clustering module 3 includes a sorting unit 30. The radar point cloud data first enters the sorting unit 30 and is sorted according to a preset priority order to obtain a point cloud data set. The step-by-step clustering module 3 also includes a level 1 clustering unit 31, a level 2 clustering unit 32, and finally an N-level clustering unit 33 divided according to the vehicle size level. In the level 1 clustering unit 31, point cloud clustering of the level 1 vehicle size range centered on the current reflection point is performed. If the clustering is completed in accordance with the requirements, the clustering process of the subsequent reflection points is performed; if it does not meet the requirements, the level 2 clustering unit 32 is continued to perform point cloud clustering of the level 2 vehicle size range centered on the current reflection point; and so on, until entering the N-level clustering unit 33 to perform point cloud clustering of the level N vehicle size range centered on the current reflection point. Finally, all clustering results are output through the clustering result output module 4.

[0072] In an embodiment of the present invention, an example is given of grading actual vehicles into small vehicles, medium vehicles and large vehicles, that is, the size of a Class 1 vehicle corresponds to a small vehicle, the size of a Class 2 vehicle corresponds to a medium vehicle, and the size of a Class 3 vehicle corresponds to a large vehicle.

[0073] Step 1. First, obtain the point cloud data Data of the current radar (the point cloud is composed of a large number of reflection points), where each reflection point has a signal strength RCS value. The larger the RCS value, the stronger the signal of the reflection point. For each reflection point, calculate the density value Density of the point (the density value is the number of points contained in a circle with the point as the center and a radius of 2.5 meters). Then sort all the reflection points in the point cloud data based on the principle of RCS value from large to small and then Density value from large to small (this sorting is to obtain the order priority of each reflection point when it is used as the core point of clustering during clustering), and obtain the sorted cloud data set Data_sorted.

[0074] Step 2: Traverse each reflection point in the sorted point cloud dataset Data_sorted in turn, regard the reflection point as the cluster core point and perform subsequent reflection point-based clustering operations on each reflection point (referring to step 2 to step 6). After the clustering operation is completed for each reflection point, the point cloud dataset Data_sorted will be updated: Step 3. First, perform a rough screening of the point cloud of the current reflection point P before clustering: filter out reflection points whose X-axis distance range from the reflection point P is in the range of [-20m, 20m] and the Y-axis distance range of [-20m, 20m] from the point cloud data set Data_sorted, and record them as the reflection point set Data_sorted_near after the initial screening; in the subsequent clustering process with the reflection point P as the clustering core, clustering points will be found from Data_sorted_near to reduce the algorithm calculation amount.

[0075] Step 4: Calculate the actual velocity direction of the reflection point P based on the X-axis velocity and Y-axis velocity of the point, and convert the velocity direction into a circular angle in the range of [0,360), recorded as A.

[0076] Step 5. Next, with the coordinates of the current reflection point P as the center and the speed direction A of the reflection point P as the direction, a rectangular frame with a length of L and a width of W is constructed. It can be considered that the rectangular frame actually represents the size of the vehicle body of the reflection point P that is currently the core point of the cluster, and subsequent clustering will also be performed within the rectangular frame.

[0077] The length and width of the rectangular box and the clustering parameters related to the rectangular box are calibrated based on the following method: ① Obtain multiple batches of radar point cloud data of traffic millimeter-wave radar in various scenarios, and filter out the point cloud data of small vehicles, medium-sized vehicles and large vehicles from the point cloud data.

[0078] ② Count the size of the point cloud and the number of reflection points contained in the point cloud for each small vehicle, medium vehicle and large vehicle (vehicle length within 6m is considered a small vehicle, between 6m-10m is considered a medium vehicle, and over 10m is considered a large vehicle).

[0079] ③ For all small vehicles, arrange the size data of their point clouds in order from small to large, and then take the size data at the 0.9 percentile as the size data of the cluster expansion rectangular box of small vehicles; medium and large vehicles determine the size data of their cluster expansion rectangular boxes in the same way.

[0080] ④ For all small vehicles, arrange the number of reflection points contained in their point clouds in ascending order, and then take the number of reflection points at the 0.8 percentile as the cluster point cloud number threshold for small vehicles; medium and large vehicles determine their cluster point cloud number threshold in the same way.

[0081] ⑤ For all small vehicles, calculate the maximum speed difference between the X-axis speed values ​​of each reflection point in each point cloud data, and arrange the maximum speed difference values ​​in order from small to large, and then take the speed difference value at the 0.9 percentile as the clustering speed difference threshold of small vehicles; medium-sized vehicles determine their clustering speed difference threshold in the same way.

[0082] Step 6: After obtaining various parameter values ​​for clustering the current reflection point P, clustering is performed.

[0083] 6-1. First, assume that the current reflection point P is the point cloud of a small vehicle. Then, the small vehicle rectangular box Rsmall can be constructed for the first time based on the speed direction A, the clustering rectangular box sizes Lsmall and Wsmall of the small vehicle; then proceed to 6-3.

[0084] 6-2. Search from the surrounding reflection points Data_sorted_near of the reflection point P based on the following conditions: a) The reflection point must be within the small vehicle rectangular frame Rsmall; b) The absolute value of the difference between the X-axis speed value of the reflection point and the X-axis speed value of the reflection point P must be smaller than the clustering speed difference threshold Ssmall for small vehicles.

[0085] 6-3. The reflection points in Data_sorted_near around the reflection point P that meet the two conditions in 6-2 can be considered as possible clustering points of the reflection point P. Then, the number N1 of all possible clustering points that meet the two conditions in 6-2 in Data_sorted_near is counted. If the number N1 is greater than the threshold value Nsmall of the number of clustering point clouds of small vehicles, it is considered that the reflection point P may be a point cloud of a medium-sized vehicle or a large vehicle, and subsequent step-by-step expansion operations are required to enter step 6-4; if the number N1 is not greater than the threshold value Nsmall of the number of clustering point clouds of small vehicles, it is considered that the reflection point P may only be a point cloud of a small vehicle or a noise point, and no subsequent expansion operations are performed. The ids of all current reflection points that meet the two conditions in 6-2 are recorded, and 6-10 is directly entered.

[0086] 6-4. Since the number of possible clustering points of the reflection point P is large, the reflection point P can be considered as the point cloud of a medium-sized vehicle. Then, the medium-sized vehicle rectangular box Rmedium can be constructed for the second time based on the speed direction A, the clustering rectangular box size Lmedium and Wmedium of the medium-sized vehicle; then proceed to 6-6.

[0087] 6-5. Search from the surrounding reflection points Data_sorted_near of the reflection point P based on the following conditions: a) The reflection point must be within the medium-sized vehicle rectangular frame Rmedium; b) The absolute value of the difference between the X-axis speed value of the reflection point and the X-axis speed value of the reflection point P must be less than the clustering speed difference threshold Smedium of medium-sized vehicles.

[0088] 6-6. The reflection points in Data_sorted_near of the surrounding reflection points of the reflection point P that meet the two conditions in 6-5 can be considered as possible clustering points of the reflection point P. Then, the number N2 of all possible clustering points that meet the two conditions in 6-5 in Data_sorted_near is counted. If the number N2 is greater than the threshold number Nmedium of the clustering point cloud of a medium-sized vehicle, it is considered that the reflection point P may be the point cloud of a large vehicle, and subsequent step-by-step expansion operations are required to enter step 6-7; if the number N2 is not greater than the threshold number Nmedium of the clustering point cloud of a medium-sized vehicle, it is considered that the reflection point P may only be the point cloud of a small or medium-sized vehicle or a noise point, and no subsequent expansion operations are performed. The ids of all current reflection points that meet the two conditions in 6-5 are recorded, and step 6-10 is directly entered.

[0089] 6-7. Since there are too many possible clustering points of the reflection point P, it can be considered that the reflection point P is a point cloud of a large vehicle. Then, the large vehicle rectangular box Rlarge can be constructed for the third time based on the speed direction A and the clustering rectangular box sizes Llarge and Wlarge of the large vehicle; then proceed to 6-9.

[0090] 6-8. Search from the surrounding reflection points Data_sorted_near of the reflection point P based on the following conditions: a) The reflection point must be located within the outer rectangular frame Rlarge of the large vehicle; b) The absolute value of the difference between the X-axis speed value of the reflection point and the X-axis speed value of the reflection point P must be less than the clustering speed difference threshold Slarge for large vehicles.

[0091] 6-9. The reflection points in the surrounding reflection points Data_sorted_near of the reflection point P that meet the two conditions in 6-8 can be considered as possible clustering points of the reflection point P. Since this embodiment currently only distinguishes the vehicle size into three levels, there is no need to count the number of possible clustering points, and no subsequent step-by-step expansion operations are performed. The IDs of all current reflection points that meet the two conditions in 6-8 are recorded, and the process goes to 6-10.

[0092] 6-10. The final clustering point based on the reflection point P is obtained by filtering the ids of the possible clustering points recorded previously from the surrounding reflection points Data_sorted_near of the reflection point P, which is recorded as Data_sorted_select. Then, the number of the final clustering points Data_sorted_select based on the reflection point P is counted, and the size of the vehicle model is identified based on the clustering point cloud number threshold Nsmall for small vehicles and the clustering point cloud number threshold Nmedium for medium-sized vehicles. That is, if the number of the final clustering points does not exceed Nsmall, the current vehicle is considered to be a small vehicle; if it exceeds Nsmall but does not exceed Nmedium, the current vehicle is considered to be a medium-sized vehicle; if it exceeds Nmedium, the current vehicle is considered to be a large vehicle.

[0093] 6-11. Next, the cluster with the reflection point P as the core is identified as a noise cluster. That is, if there is only one cluster point in the end, the current cluster with the reflection point P as the core is considered to be a noise cluster, otherwise it is not.

[0094] 6-12. Based on the id of the reflection point, the current reflection point P and its final cluster point Data_sorted_select are removed from the total point cloud data set Data_sorted.

[0095] 6-13. Return the current cluster with the reflection point P as the core, the vehicle size identifier of the cluster, and the identifier of whether it is a noise cluster, and return the point cloud dataset Data_sorted after removing the data of the current cluster with the reflection point P as the core; then return to step 2 to start the clustering operation of the next reflection point.

[0096] Step 7: After traversing all the reflection points in the point cloud dataset Data_sorted, for each cluster with different reflection points as the core, if it is marked as a noise cluster, it will be discarded; for each cluster not marked as a noise cluster, these clusters and their vehicle size identifiers are returned, and this clustering is completed.

[0097] The above embodiments are further elaborations and illustrations of the present invention for ease of understanding, and are not limitations of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A radar point cloud clustering method based on vehicle size adaptive discrimination, characterized in that: include: Set the parameters for the size range of each level of vehicle models during clustering, as well as the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; Acquire radar point cloud data and sort them to obtain a point cloud data set; The reflection points in the point cloud data set are traversed in order to perform clustering based on the reflection points, and the point cloud clustering with the vehicle size range gradually expanded with the current reflection point as the center is performed to obtain a cluster cluster with the current reflection point as the cluster core; Screening out the point cloud data in the cluster from the point cloud data set, and continuing to perform clustering based on reflection points; After traversing the point cloud dataset, several clusters and corresponding vehicle model sizes are obtained.

2. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 1 is characterized in that: The step of acquiring radar point cloud data and sorting the point cloud data set comprises: The signal strength of the reflection points in the point cloud data is ranked from large to small as the first priority; The density values ​​of the reflection points are ranked from large to small as the second priority.

3. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 1 or 2, characterized in that: The clustering based on reflection points includes: setting the size range of vehicle models from level 1 to level N from small to large; When clustering the point cloud of the i-level vehicle size range with the current reflection point as the center, the number of point clouds Mi of the reflection points that meet the i-level vehicle size judgment standard is selected; If Mi is greater than the clustering point cloud quantity threshold of the i-level vehicle size range, point cloud clustering of the next level vehicle size range is performed; otherwise, a cluster cluster with the current reflection point as the clustering core and the corresponding i-level vehicle size is obtained.

4. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 3 is characterized in that: The point cloud clustering of the i-level vehicle size range with the current reflection point as the center includes: Taking the coordinates of the current reflection point as the center and the speed direction of the current reflection point as the length direction, a rectangular frame with a length of Li and a width of Wi is constructed to obtain the size range of the i-level vehicle model at the current reflection point; Point cloud clustering is performed within the i-class vehicle size range of the current reflection point.

5. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 3 is characterized in that: The i-class vehicle size judgment criteria include: The reflection point is within the i-class vehicle size range of the current reflection point; and The absolute value of the speed difference between the reflection point and the current reflection point is less than the cluster speed difference threshold of the i-class vehicle size range; The speed difference is a difference in projection speed in a direction in which the road extends.

6. A radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 1 or 2 or 4 or 5, characterized in that: The parameters for the size range of each level of vehicle models when setting clustering include: Obtain historical radar point cloud data and filter out point cloud data of each level of vehicle size; for level i vehicle size, obtain the size of all point clouds of level i vehicle size and the number of reflection points contained in each point cloud; Arrange the sizes of all point clouds in descending order, and select the size at the size setting percentile as the parameter of the size range of the i-level vehicle model.

7. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 6 is characterized in that: The number of reflection points contained in each point cloud based on the i-level vehicle size is sorted in descending order, and the number of reflection points at the set number quantile is selected as the cluster point cloud number threshold for the i-level vehicle size range; The maximum speed difference between the reflection points in each point cloud based on the i-level vehicle size is sorted in descending order, and the maximum speed difference at the speed setting percentile is selected as the clustering speed difference threshold of the i-level vehicle size range; The speed difference is a difference in projection speed in a direction in which the road extends.

8. The radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 1, 2, 4 or 5, characterized in that: The clustering based on reflection points also includes initial screening of point clouds; Set the point cloud initial screening range centered on the current reflection point, where the point cloud initial screening range includes the maximum range of the size range of each level of vehicle models; Filter out a set of reflection points within the initial screening range of the point cloud from the radar point cloud data; Point cloud clustering is performed based on the reflection point set with a gradually expanding vehicle size range.

9. A radar point cloud clustering method based on vehicle size adaptive discrimination according to claim 1 or 2 or 4 or 5, characterized in that: Perform noise clustering judgment on the clusters that have completed clustering: When the number of reflection points in a cluster is less than the noise point threshold, it is judged as a noise cluster; After traversing the point cloud data set, the clusters judged as noise clusters are discarded, and several clusters and corresponding vehicle model sizes are obtained.

10. A radar point cloud clustering system based on adaptive vehicle size discrimination, applicable to the radar point cloud clustering method according to any one of claims 1 to 9, characterized in that: include: The historical point cloud saving module saves historical radar point cloud data, sets the parameters of the size range of each level of vehicle models, and sets the corresponding cluster point cloud quantity threshold and cluster speed difference threshold; A radar point cloud acquisition module is used to acquire radar point cloud data to be clustered; The step-by-step clustering module traverses the reflection points in the point cloud dataset to perform clustering based on the reflection points, and performs point cloud clustering with a gradually expanding vehicle size range with the current reflection point as the center.

Images