Millimeter wave radar adaptive clustering method, data processing equipment and storage medium

Abstract

The millimeter-wave radar adaptive clustering method, data processing equipment and storage medium cluster the point cloud data acquired by the vehicle-mounted millimeter-wave radar. First, the point cloud data is screened and dynamic point traces are selected as the data to be clustered; secondly, the neighborhood threshold is adaptively set, and the absolute value of the radial distance difference between the point traces is compared with the neighborhood threshold, and the point traces that are not in the neighborhood of the current point trace are pre-filtered to further reduce the amount of calculation during clustering; finally, according to the uneven density characteristics of the millimeter-wave radar point cloud, the relationship between the distance, distance resolution and angle resolution based on the point trace is used as the neighborhood distance threshold for clustering, and the obtained clustering results are used for target tracking. The present invention effectively improves the clustering effect for millimeter-wave radar point cloud data with large data volume and uneven density, which is beneficial to subsequent target tracking tasks.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle-mounted millimeter-wave radars, relates to data processing of vehicle-mounted millimeter-wave radars, and is a millimeter-wave radar adaptive clustering method, data processing equipment and storage medium. Background Art

[0002] As one of the important sensors for advanced driver assistance systems and unmanned driving, millimeter-wave radar plays an important role in improving traffic safety factors and reducing the incidence of traffic accidents. The target tracking module is one of the key parts of the vehicle-mounted millimeter-wave radar system and an important part of determining the performance of the radar. In practical applications, in order to distinguish targets, the point cloud acquired by the radar will be clustered first, the clustering centroid of the point cloud set will be obtained, and then subsequent target tracking will be performed based on the clustering results. The clustering of millimeter-wave radar point clouds lays the foundation for target tracking and has a very significant effect on the performance of target tracking.

[0003] Traditional clustering methods include the K-means algorithm and the density-based spatial clustering of noise applications (DBSCAN) algorithm. The K-means algorithm requires the number of clusters to be known in advance and cannot meet real-time performance requirements. The DBSCAN algorithm is a density-based method that uses the set neighborhood radius and minimum number of points as density thresholds. If a point's neighborhood radius contains at least the minimum number of points, the point is considered a core point. The DBSCAN algorithm does not require a preset number of clusters and can cluster targets of any shape. It also has the advantage of removing noise points and has been widely used in the millimeter-wave radar field. However, with the substantial increase in the number of point clouds brought about by the improvement of the millimeter-wave radar's resolution, the DBSCAN algorithm has the problem of low efficiency on the one hand because it has to traverse the point cloud. On the other hand, in the target tracking scenario, the millimeter-wave point cloud data has an uneven density. The use of a unified density threshold will cause distant targets to be treated as noise, affecting the tracking of subsequent targets. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in the millimeter-wave radar point cloud data processing clustering module, with the improvement of the accuracy of millimeter-wave radar, the number of point clouds increases day by day, and the data processing efficiency of the existing clustering method is low; at the same time, the density of near and far targets of the millimeter-wave radar is uneven, and traditional clustering algorithms such as DBSCAN perform clustering based on a unified density threshold, which cannot effectively cope with the complex situations in actual target tracking applications, and easily lead to the failure of clustering distant targets or the problem of multiple nearby targets being clustered together, thereby affecting subsequent target tracking.

[0005] The technical solution of the present invention is: a millimeter wave radar adaptive clustering method, which clusters the point cloud data acquired by the vehicle-mounted millimeter wave radar for target distinction in subsequent target tracking, and the clustering includes the following steps:

[0006] 1) Filter the point cloud data, filter out the point data with a signal-to-noise ratio lower than the set threshold, select dynamic points according to the point speed information and the vehicle speed information, and obtain the input point set for clustering. Data ;

[0007] 2) Point trace set Data Adaptive clustering;

[0008] 2.1) Set the adaptive neighborhood threshold Eps 2 And the clustering point threshold MinPoints, for points p1 and p2, let r1 and r2 be the radial distances of points p1 and p2, θ1 and θ2 be the azimuths of points p1 and p2, Δr and Δθ represent the radial distance difference and azimuth angle difference of points p1 and p2, respectively. According to the Euclidean distance of the two points:

[0009]

[0010] The radial distance difference Δr between two points of the same target and the detected distance resolution r res Related, the azimuth angle difference Δθ and the detection angle resolution θ res Related, let Δr = w r r res , Δθ = w θ θ res ,in, w r is Δr in r res The weighted value on w θ is Δθ at θ res The weighted value on

[0011]

[0012] Set the neighborhood threshold accordingly , that is, the neighborhood threshold gate changes adaptively with the distance between points.

[0013] The clustering point threshold MinPoints is set in multiple segments according to the radial distance, and different clustering point thresholds are set in different radial distance segments;

[0014] 2.2) Traverse the point set Data The points in the are clustered preliminarily according to the neighborhood threshold Eps 2And the clustering point threshold MinPoints, judge Data Whether other points are in the neighborhood of the current point p1, and whether the number of points in the neighborhood is greater than MinPoints according to the segment statistics of the radial distance:

[0015] If the MinPoints condition is met, p1 is the core point, a new cluster A is created, and all points in the neighborhood of p1 are classified into cluster A. At the same time, a new queue B is created accordingly, and all points in the neighborhood of point p1 are added to queue B;

[0016] If the MinPoints condition is not met, then the point p1 is not a core point and is marked as noise. Get the next point. If the point has been marked as a cluster or marked as noise, continue to get the next point until you get a point that is not marked, and determine whether it is a core point or noise.

[0017] 2.3) Traverse queue B. If the acquired point is marked as noise, continue to acquire the next point in queue B until the acquired point is not marked as noise. At this time, the acquired point is recorded as p3. It is determined whether point p3 is a core point. If it is a core point, all points in the neighborhood of point p3 are added to queue B and classified into cluster A until all points in queue B are traversed;

[0018] 3) Traverse the point set Data Get the final clustering result.

[0019] Furthermore, in step 1), the selection of dynamic points is as follows: judging whether the radial speed after removing the influence of the vehicle speed from the points is greater than a set threshold, and the formula is:

[0020]

[0021] in radv is the radial velocity of the point, egoV is the vehicle speed, azimuth is the point azimuth, threshold For the set dynamic point velocity threshold, select the points that meet the formula conditions as clustering input.

[0022] Furthermore, in step 2.1), the cluster point threshold MinPoints is set in multiple segments according to the radial distance. p have:

[0023]

[0024] in, range _ i is the threshold of distance segmentation, num i is the cluster point threshold at different distances,i Indicates the segment number. r For dot trace p radial distance.

[0025] Furthermore, in step 2.2), the point that is not in the neighborhood of point p1 is initially screened by using the method that the absolute value of the radial distance difference between the points is greater than the neighborhood value Eps, that is, , then it is considered that point p2 is not in the neighborhood of point p1, and then the remaining points are calculated according to the neighborhood threshold Eps 2 Determine whether it is within the neighborhood of the current point p1.

[0026] The present invention also provides a data processing device for adaptive clustering of millimeter-wave radars, comprising a data processing module, wherein at least one instruction or at least one program is loaded into the data processing module, and the data processing module receives point cloud data and vehicle motion information acquired by a vehicle-mounted millimeter-wave radar. When the at least one instruction or the at least one program is executed, the above-mentioned adaptive clustering method is implemented on the received point cloud data, and the clustering result of the point cloud data is output.

[0027] The present invention also provides a computer-readable storage medium, which stores at least one instruction or at least one program. When the at least one instruction or at least one program is executed, the above-mentioned adaptive clustering method is implemented on the point cloud data based on the point cloud data and vehicle motion information acquired by the vehicle-mounted millimeter-wave radar.

[0028] The present invention proposes an efficient adaptive clustering method based on the characteristics of millimeter wave radar point cloud data. First, the point cloud data is screened, and the point data with low signal-to-noise ratio is filtered out. Dynamic point data is selected according to the point information and vehicle information as the input of the clustering module; secondly, the point that is not in the neighborhood is initially screened by using the method that the absolute value of the radial distance difference between the point data is greater than the neighborhood distance threshold, which can further reduce the amount of program calculation; thirdly, according to the uneven density of millimeter wave radar point cloud, the relationship between the distance, distance resolution and angle resolution based on the point data is used as the neighborhood distance threshold for adaptive clustering, which improves the traditional clustering due to the use of a unified threshold for clustering, resulting in distant targets not being clustered or near targets appearing as false clustered targets, thereby effectively improving the clustering effect. The experimental results also prove that the present invention effectively improves the clustering effect, which is beneficial to subsequent target tracking tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a framework diagram of the method of the present invention.

[0030] Figure 2 The figure is a detailed flow chart of the method of the present invention.

[0031] Figure 3Schematic diagram of clustering results of the prior art DBSCAN algorithm in the embodiment.

[0032] Figure 4 Schematic diagram of clustering results of the method of the present invention in the embodiment.

[0033] Figure 5 The figure is a comparison chart of clustering time between traditional DBSCAN and the method of the present invention in the embodiment. DETAILED DESCRIPTION

[0034] The specific implementation of the present invention is described below with reference to the accompanying drawings.

[0035] like Figure 1 As shown in FIG. 1 , the present invention performs efficient adaptive clustering based on the motion information of the vehicle and the point cloud data obtained by the millimeter wave radar, so as to distinguish the targets in the subsequent target tracking. The adaptive clustering method of the present invention first screens the points to reduce the amount of data processing, and then performs adaptive clustering. The flowchart is shown in FIG. Figure 2 , the specific implementation is as follows.

[0036] Step 1: Filter points.

[0037] Step 1.1: Set the signal-to-noise ratio threshold Th , filter out signal-to-noise ratios below Th The point trace data.

[0038] Step 1.2: Select dynamic points according to the point information and vehicle information as the point input Data of the clustering module.

[0039] The dynamic point trace is selected based on whether the radial speed after removing the influence of the vehicle speed is greater than the set threshold. The judgment formula is:

[0040]

[0041] in radv is the radial velocity of the point, egoV is the vehicle speed, azimuth is the point azimuth, threshold To set the dynamic point velocity threshold, select a value greater than the threshold. threshold The points are taken as clustering objects.

[0042] Step 2: Efficient adaptive clustering.

[0043] Step 2.1: Set the adaptive neighborhood threshold Eps 2 And the clustering point threshold MinPoints.

[0044] In point clustering, the coordinates of points p1 and p2 in the vehicle coordinate system are assumed to be and , the Euclidean distance between points p1 and p2 is , if satisfied , then the point p2 is considered to be in the neighborhood of the point p1.

[0045] The present invention is studied on this basis. For the point traces p1 and p2, let r1 and r2 be the radial distances of the point traces p1 and p2, θ1 and θ2 be the azimuths of the point traces p1 and p2, Δr and Δθ represent the radial distance difference and azimuth angle difference of the point traces p1 and p2, respectively. Then:

[0046]

[0047] make ,

[0048] Then we have:

[0049]

[0050] The radial distance difference Δr between two points of the same target is related to the detection distance resolution r res Related, the azimuth angle difference Δθ and the detection angle resolution θ res Related, let Δr = w r r res , Δθ = w θ θ res ,in, w r is Δr in r res The weighted value on w θ is Δθ at θ res The weighted value on

[0051] .

[0052] Based on this, the present invention sets an adaptive neighborhood threshold , that is, the neighborhood threshold gate changes adaptively with the distance between points.

[0053] At the same time, the present invention sets the cluster point count threshold MinPoints in the neighborhood according to the radial distance in multiple segments, and different radial distance segments have different cluster point count thresholds. p have:

[0054]

[0055] in, range _ i is the threshold of distance segmentation, num iis the cluster point threshold at different distances, i Indicates the segment number. r For dot trace p radial distance.

[0056] Taking the two-segment method as an example, the point count threshold MinPoints in the core point neighborhood is set according to the distance segmentation, as follows:

[0057]

[0058] in, range _1 is the distance threshold, num 1, num 2 is the cluster point threshold value at different distances. range All points within _1, num 1. Determine whether the number of cluster points meets the requirements of core points, and check whether the radial distance is greater than or equal to range All points of _1, num 2. Determine whether the number of clustering points meets the requirements of core points.

[0059] Step 2.2: Traverse the point set Data The points in the are clustered preliminarily according to the neighborhood threshold Eps 2 And the clustering point threshold MinPoints, first determine Data Check whether other points are in the neighborhood of the current point p1, and count whether the number of points in the neighborhood of p1 is greater than the corresponding MinPoints.

[0060] If the number of points in each segment in the neighborhood of point p1 is greater than the clustering point threshold MinPoints, then p1 is the core point, a new cluster A is created, and all points in the neighborhood of p1 are classified into cluster A. At the same time, a new queue B is created, and all points in the neighborhood of point p1 except p1 are added to queue B.

[0061] Furthermore, since this step is time-consuming to traverse the point traces, the present invention preferably uses a method in which the absolute value of the radial distance difference between the point traces is greater than the neighborhood value Eps to preliminarily screen out the points that are not in the neighborhood of the point trace p1, that is, , then it is considered that point p2 is not in the neighborhood of point p1.

[0062] If the number of points in each distance segment within the neighborhood of point p1 is less than the corresponding cluster point threshold MinPoints, then point p1 is not a core point and is marked as noise. Get the next point. If the point has been marked as a cluster or marked as noise, continue to get the next point until an unmarked point is obtained to determine whether it is a core point or noise.

[0063] Step 2.3: Cluster expansion.

[0064] Traverse queue B. If the acquired point trace has been marked as noise, continue to acquire the next point in queue B until the acquired point trace is not marked as noise. At this time, the acquired point trace is recorded as p3.

[0065] Determine whether point p3 is a core point. If it is a core point, add all points in the neighborhood of point p3 (except point p3) to queue B and assign them to cluster A until all points in queue B are traversed.

[0066] Finally, if all points in Data are traversed, clustering is completed.

[0067] The method of reducing the amount of data in the present invention includes filtering out points with low signal-to-noise ratio, selecting dynamic points, and using the method that the absolute value of the radial distance difference between points is greater than the neighborhood distance threshold to preliminarily screen out points that are not in the neighborhood. 2 ) algorithm, filtering out points with low signal-to-noise ratio and selecting dynamic points to reduce n from the source. Secondly, the point that is not in the neighborhood is initially screened out by using the method that the absolute value of the radial distance difference between points is greater than the neighborhood distance threshold. As the number of traversals increases, the calculation time can also be reduced.

[0068] The embodiment of the present invention also provides a data processing device for adaptive clustering of millimeter wave radar, including a data processing module, in which at least one instruction or at least one program is loaded, the data processing module receives point cloud data and vehicle motion information acquired by the vehicle-mounted millimeter wave radar, and when the at least one instruction or the at least one program is executed, the above-mentioned adaptive clustering method is implemented for the received point cloud data, and the clustering result of the point cloud data is output. The data processing device of the embodiment of the present invention can be represented as various forms of digital electronic computer devices, such as notebook computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, large computers, and other suitable computers, and can also represent various forms of mobile devices, such as personal digital processing terminals, smart phones, wearable devices and other similar data computing devices. The data processing module can be various general and / or special processing components with processing and computing capabilities, some examples include but are not limited to central processing units (CPUs), graphics processing units (GPUs), various special artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc.

[0069] The embodiment of the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program is used to implement the method according to the embodiment of the present invention when executed. The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0070] The effects of the present invention are embodied below through embodiments.

[0071] In this embodiment, the signal-to-noise ratio threshold Th is set to 10 dB for filtering out the points with low signal-to-noise ratio, and the dynamic point speed threshold threshold Set to 0.5, adaptively set the neighborhood threshold Eps 2 middle w r Set to 3, w θ Set to 3, MinPoints is divided into two segments, segment threshold range _1 is set to 80, num 1 is set to 5, num 2 is set to 3. See the comparison of clustering effects for details. Figure 3 and Figure 4 , Figure 3 In order to use the traditional DBSCAN clustering results, since the distant target points are relatively scattered compared to the nearby target points, the traditional DBSCAN clustering uses a fixed and unified Eps 2 and MinPoints threshold, resulting in unsuccessful clustering of targets in the distance. Figure 4 This is the efficient adaptive clustering result proposed by the present invention, and all targets are clustered successfully. Figure 5 This is a time comparison chart of using traditional DBSCAN clustering and using the efficient adaptive clustering solution of the present invention. It can be seen that the program running efficiency of the present invention is significantly improved.

Claims

1. A millimeter wave radar adaptive clustering method, characterized by: The point cloud data acquired by the vehicle-mounted millimeter-wave radar is clustered for target differentiation in subsequent target tracking, and the clustering includes the following steps: 1) Filter the point cloud data, filter out the point data with a signal-to-noise ratio lower than the set threshold, select dynamic points according to the point speed information and the vehicle speed information, and obtain the input point set for clustering. Data ; 2) Point trace set Data Adaptive clustering; 2.1) Set the adaptive neighborhood threshold Eps 2 And the clustering point threshold MinPoints, for points p1 and p2, let r1 and r2 be the radial distances of points p1 and p2, θ1 and θ2 be the azimuths of points p1 and p2, Δr and Δθ represent the radial distance difference and azimuth angle difference of points p1 and p2, respectively. According to the Euclidean distance of the two points: The radial distance difference Δr between two points of the same target and the detected distance resolution r res Related, the azimuth angle difference Δθ and the detection angle resolution θ res Related, let Δr = w r r res , Δθ = w θ θ res ,in, w r is Δr in r res The weighted value on w θ is Δθ at θ res The weighted value on Set the neighborhood threshold accordingly , that is, the neighborhood threshold gate changes adaptively with the distance between points. The clustering point threshold MinPoints is set in multiple segments according to the radial distance, and different clustering point thresholds are set in different radial distance segments; 2.2) Traverse the point set Data The points in the are preliminarily clustered according to the neighborhood threshold Eps 2 And the clustering point threshold MinPoints, judge Data Whether other points are in the neighborhood of the current point p1, and whether the number of points in the neighborhood is greater than MinPoints according to the segment statistics of the radial distance: If the MinPoints condition is met, p1 is the core point, a new cluster A is created, and all points in the neighborhood of p1 are classified into cluster A. At the same time, a new queue B is created accordingly, and all points in the neighborhood of point p1 are added to queue B; If the MinPoints condition is not met, then the point p1 is not a core point and is marked as noise. Get the next point. If the point has been marked as a cluster or marked as noise, continue to get the next point until you get a point that is not marked, and determine whether it is a core point or noise. 2.3) Traverse queue B. If the acquired point is marked as noise, continue to acquire the next point in queue B until the acquired point is not marked as noise. At this time, the acquired point is recorded as p3. It is determined whether point p3 is a core point. If it is a core point, all points in the neighborhood of point p3 are added to queue B and classified into cluster A until all points in queue B are traversed; 3) Traverse the point set Data Get the final clustering result.

2. The millimeter wave radar adaptive clustering method according to claim 1, characterized in that: In step 1), the selection of dynamic points is as follows: judging whether the radial speed after removing the influence of the vehicle speed from the points is greater than the set threshold, and the formula is: in radv is the radial velocity of the point, egoV is the vehicle speed, azimuth is the point azimuth, threshold For the set dynamic point velocity threshold, select the points that meet the formula conditions as clustering input.

3. The millimeter wave radar adaptive clustering method according to claim 1, characterized in that: In step 2.1), the cluster point threshold MinPoints is set in multiple segments according to the radial distance. p have: in, range _ i is the threshold of distance segmentation, num i is the cluster point threshold at different distances, i Indicates the segment number. r For dot trace p radial distance.

4. The millimeter wave radar adaptive clustering method according to claim 1, characterized in that: In step 2.2), the points that are not in the neighborhood of point p1 are initially screened by using the method that the absolute value of the radial distance difference between the points is greater than the neighborhood value Eps, that is, , then it is considered that point p2 is not in the neighborhood of point p1, and then the remaining points are calculated according to the neighborhood threshold Eps 2 Determine whether it is within the neighborhood of the current point p1.

5. A data processing device for adaptive clustering of millimeter wave radar, characterized in that: The method comprises a data processing module, wherein at least one instruction or at least one program is loaded in the data processing module, and the data processing module receives point cloud data and vehicle motion information acquired by a vehicle-mounted millimeter-wave radar. When the at least one instruction or the at least one program is executed, the adaptive clustering method according to any one of claims 1 to 4 is implemented on the received point cloud data, and a clustering result of the point cloud data is output.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction or at least one program. When the at least one instruction or at least one program is executed, the adaptive clustering method as described in any one of claims 1 to 4 is implemented on the point cloud data based on the point cloud data and vehicle motion information obtained by the on-board millimeter-wave radar.

Images