A method for quickly extracting laser radar ground point cloud

Abstract

The present invention relates to a method for quickly extracting laser radar ground point clouds, comprising the following steps: real-time acquisition of high-precision positioning information for characterizing the current position of a vehicle; conversion of point clouds from a laser radar coordinate system to a vehicle body coordinate system; real-time generation of a road ROI area point set of the current vehicle's self-positioning information; generation of a contour area based on the road ROI area point set; generation of multiple small areas based on the road ROI area point set; a contour area is a collection of multiple small areas; a ground point cloud data set and a non-ground point cloud data set are obtained based on the contour area and the small areas; a ground point cloud data set is output; and the ground point cloud data set is the final result of the present invention. The present invention solves the difficulty of extracting ground points with large curvature and slope; does not rely on data sample annotation, greatly reducing costs and computing power requirements; does not require segmentation through relationship mapping between different sensors, does not increase dependencies, and does not require higher requirements for calibration accuracy and time synchronization.

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving technology, and in particular to a method for quickly extracting laser radar ground point clouds. Background Art

[0002] Environmental perception technology is the most important module of intelligent driving, and accurate environmental perception technology is particularly important. Currently, LiDAR is one of the essential sensors for high-level autonomous driving, so how to process LiDAR data is particularly important. Among them, effectively segmenting or extracting ground points from the point cloud scanned by the laser sensor is the most important preprocessing link for target perception and recognition.

[0003] Currently, laser point cloud ground point extraction is mainly divided into the following two technical paths:

[0004] 1. According to the spatial position of the ground point cloud, an algorithm is designed to extract the ground point cloud through geometric constraints.

[0005] A typical example is paper 1 "Fast segmentation of 3D point clouds: A paradigm on LiDAR data for autonomous vehicle applications", which designs the main steps of ground point cloud extraction as follows:

[0006] (1) First, the point cloud is sorted according to the elevation z to obtain the preliminary point cloud within the range of the highest ground point.

[0007] (2) After obtaining the initial set of ground point clouds, the ground equation is obtained through SVD (singular value decomposition).

[0008] (3) Finally, the final ground point set is obtained by iterating N (setting parameters) times.

[0009] A typical example is paper 2 "Fast segmentation of 3d point clouds for ground vehicles", in which the main steps of extracting ground point clouds are:

[0010] (1) First, the point cloud area is divided into several sector-shaped areas, and each sector-shaped area is further divided into sub-areas.

[0011] (2) Connect the points of each sub-region of each fan-shaped region. If the connection slope between the sub-region point inside the fan-shaped region and the adjacent sub-region point is lower than a certain set parameter threshold, it is a ground point.

[0012] (3) Repeat the same operation in step (2) for each sector area to obtain the final ground point. The main operation of the spatial geometric constraint method to extract the ground point cloud is to divide the grid, search the grid for the lowest point set, fit the iterative generation surface to finally obtain the ground point.

[0013] 2. Mainly through deep learning to segment ground point clouds. Among them: deep learning point cloud segmentation algorithms mainly include the following methods:

[0014] 2.1 Extract ground points by performing point cloud segmentation directly on the original point cloud structure or point cloud projection through deep learning.

[0015] A typical example is point cloud segmentation represented by Paper 3 “RangeNet++: Fast and Accurate LiDAR Semantic Segmentation”, whose main steps are: point cloud projection, convolutional neural network semantic segmentation, point cloud reconstruction, and post-processing module.

[0016] 2.2 is to accelerate the segmentation by projecting the point cloud into the image with the help of visual methods. Regardless of which deep learning strategy is used to extract ground point clouds, the effect of ground point cloud extraction is getting better and better with the improvement of data annotation quality and data volume.

[0017] The defects of the prior art are:

[0018] 1. The point cloud geometric constraint removal algorithm is time-consuming and complex, and cannot adapt to scenes with large changes in ground slope;

[0019] 2. Since the point cloud geometric constraint removal algorithm needs to adjust parameters according to the scene, the scene adaptability is poor;

[0020] 3. Since the point cloud deep learning ground segmentation algorithm is time-consuming and has high labeling costs, this method has certain difficulties in industrial implementation. Summary of the invention

[0021] In view of the above problems, the present invention provides a method for quickly extracting lidar ground point clouds, which aims to solve the difficulty in extracting ground points with large curvature and slope; it does not rely on data sample annotation, greatly reducing costs and computing power requirements; it does not require segmentation through relationship mapping between different sensors, and does not increase dependencies, and does not require higher requirements for calibration accuracy and time synchronization.

[0022] To solve the above problems, the technical solution provided by the present invention is:

[0023] A method for quickly extracting laser radar ground point clouds includes the following steps:

[0024] S100. Acquire high-precision positioning information for characterizing the current position of the vehicle in real time; then convert the point cloud from the laser radar coordinate system to the vehicle body coordinate system according to the high-precision positioning information;

[0025] S200. Generate a road ROI area point set of the current vehicle's self-positioning information in real time; the road ROI area point set is located under the vehicle body coordinate system;

[0026] S300. Generate a contour area according to the road ROI area point set; and generate multiple small areas according to the road ROI area point set; the contour area is a collection of multiple small areas;

[0027] S400. Obtaining a ground point cloud data set and a non-ground point cloud data set according to the contour area and the small area;

[0028] S500. Output the ground point cloud dataset.

[0029] Preferably, in S100, the high-precision positioning module calculates, updates and sends the high-precision positioning information in real time;

[0030] The laser radar software algorithm module receives the high-precision positioning information in real time and converts the point cloud from the laser radar coordinate system to the vehicle body coordinate system.

[0031] Preferably, S200, after receiving the high-precision positioning information, the high-precision map generates and dynamically sends the road ROI area point set where the current vehicle is located in real time according to the high-precision positioning information;

[0032] The points in the road ROI area point set include abscissa, ordinate and elevation coordinates;

[0033] The points in the road ROI area point set are taken from a static point set; the static point set is converted to the vehicle body coordinate system based on high-precision positioning information, and is obtained and dynamically sent by the high-precision map based on a pre-built-in priori layer combined with the high-precision positioning information.

[0034] Preferably, generating a contour area according to the road ROI area point set in S300 specifically includes the following steps:

[0035] S310a. Receive the road ROI area point set sent by the high-precision map; then fit the road ROI area point set into a polygon closed area; the points in the polygon closed area only take the XY coordinates of each point, including the horizontal coordinate and the vertical coordinate; the polygon closed area is a two-dimensional polygonal plane area;

[0036] S320a. Filter the point cloud data generated by the real-time scanning of the laser radar through the polygon closed area; the filtering rule only takes the XY coordinates of each point, filters according to the XY plane, and does not restrict z; then retain the point cloud inside the polygon closed area of ​​the polygon to obtain the contour area.

[0037] Preferably, the step of generating a plurality of small areas according to the road ROI area point set in S300 specifically includes the following steps:

[0038] S310b receiving the road ROI area point set sent by the high-precision map;

[0039] S320b. Dynamically generate a plurality of the small areas according to the road ROI area point set; the coordinates of the vertices of each small area are three-dimensional points, including a horizontal coordinate, a vertical coordinate, and an elevation coordinate.

[0040] Preferably, S400 specifically includes the following steps:

[0041] S410. Receive the small area obtained in S310b to S320b; then use the vertices of each small area to generate a corresponding three-dimensional cutting plane for each small area; the three-dimensional cutting plane is a surface structure in a three-dimensional coordinate system;

[0042] S420. Loop through each of the three-dimensional cutting planes, and determine the corresponding three-dimensional cutting planes to which the point clouds in the polygon closed area belong in turn; then divide the point clouds in the polygon closed area into each of the corresponding three-dimensional cutting planes;

[0043] S430. According to the distance from each point to the three-dimensional tangent plane, determine whether the point is above or below the three-dimensional tangent plane; then perform the following operations according to the determination result:

[0044] If the point is above the three-dimensional tangent plane, the point is marked as a non-ground point;

[0045] If the point is below the three-dimensional tangent plane, the point is marked as a ground point;

[0046] S440. Pack all the non-ground points to obtain the non-ground point cloud data set in the polygon closed area; pack all the ground points to obtain the ground point cloud data set in the polygon closed area.

[0047] Preferably, generating a corresponding three-dimensional cutting plane for each small area in S410 specifically includes the following steps:

[0048] S411. Real-time dynamic receiving of the road ROI area point set;

[0049] S412. Dividing the road ROI area point set into a plurality of small areas of different sizes;

[0050] S413. Loop through each of the small areas; solve the plane equation according to the vertex coordinates to obtain the three-dimensional tangent plane; the three-dimensional tangent plane is expressed as follows:

[0051] Ax+By+Cz+D=0

[0052] Among them: x is the horizontal coordinate; y is the vertical coordinate; z is the elevation coordinate; (A, B, C) is the normal vector; D is the distance from the origin to the plane.

[0053] Preferably, solving the plane equation in S413 specifically includes the following steps:

[0054] S413a. Obtain the number of vertices of the small area corresponding to the plane equation to be solved;

[0055] S413b. Perform the following operations according to the number of vertices in the small area:

[0056] If the number of vertices in the small area is greater than 3, the least square method is used to solve the plane equation;

[0057] If the number of vertices in the small area is 3, the solution of the overdetermined equation is directly obtained through the vertex information to solve the plane equation.

[0058] Preferably, the step of dividing the road ROI area point set into a plurality of small areas of different sizes in S412 specifically comprises the following steps:

[0059] S412a traverses all points in the road ROI area point set sent by the high-precision map;

[0060] S412b. Sorting all points in S412a in the X-axis direction from the longitudinal distance of the vehicle according to the rule from near to far to form a sorted point set;

[0061] S412c. The sorted point set is connected into a closed sub-region every 4 points; and the generated closed sub-region is interpreted as a small region;

[0062] S412d. According to the number of points remaining in the sorted point set, perform the following operations:

[0063] If the number of remaining points in the sorted point set is greater than 3, return and execute S412c again;

[0064] If the number of remaining points in the sorted point set is equal to 3, then these 3 points are connected to form a closed sub-region; then the generated closed sub-region is interpreted as one of the small regions;

[0065] If the number of remaining points in the sorted point set is less than 3, the last remaining point is discarded; then the process of S412 ends.

[0066] Preferably, in S412c, every 4 connections form a closed sub-region, and the rule is: 2 points on each of the left and right edges.

[0067] Compared with the prior art, the present invention has the following advantages:

[0068] 1. The present invention optimizes the existing spatial geometric constraint algorithm to adaptively extract dynamic ground point clouds, which has the disadvantages of high algorithm time consumption, poor scene adaptability and the need for a large amount of manual parameter adjustment. It also solves the problem of difficulty in extracting ground points with large curvature and slope;

[0069] 2. Since the present invention optimizes the existing deep learning segmentation and extraction method of ground points, and directly segment and extract the lidar point cloud, it does not rely on data sample annotation, greatly reducing data costs and computing power requirements;

[0070] 3. Since the present invention does not rely on the segmentation and extraction of the lidar point cloud, and since the current laser point cloud segmentation method has high requirements on computing power and cannot be actually used in the industry, there is no need to segment through relationship mapping between different sensors, and there is no need to increase dependencies and no higher requirements for calibration accuracy and time synchronization. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic diagram of the overall algorithm flow of a specific embodiment of the present invention;

[0072] Figure 2 A schematic diagram of the entire process of visually displaying ground point extraction according to a specific embodiment of the present invention;

[0073] Figure 3a Schematic diagram of the point cloud of interest in the ROI area compared with the traditional GPF algorithm;

[0074] Figure 3b A schematic diagram of a point cloud of interest in a ROI region of contrast effect according to a specific embodiment of the present invention;

[0075] Figure 4a This is a schematic diagram of the effect of GPF extracting ground point cloud;

[0076] Figure 4bThis is a schematic diagram of the effect of extracting ground point cloud according to a specific embodiment of the present invention; DETAILED DESCRIPTION

[0077] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0078] It should be explained in advance that the method for quickly extracting lidar ground point clouds proposed in the present invention mainly optimizes the following aspects of the spatial geometric constraint method. First, the method proposed in the present invention is less time-consuming than the spatial geometric constraint in terms of algorithm. Secondly, relying on high-precision maps to provide road edge information, the scene adaptability of the algorithm of the present invention is improved. Finally, based on the road edge information provided by high-precision maps, the algorithm proposed in the present invention further solves the extraction of ground point clouds on slopes with large curvature by designing an algorithm flow. On the whole, the lidar ground point cloud extraction algorithm proposed in the present invention has a clear process, low time consumption, and good scene adaptability, while solving the difficulty of cleanly extracting ground point clouds on slopes with large curvature.

[0079] The present invention proposes a method for quickly extracting laser radar ground point clouds, which mainly optimizes the following aspects of the spatial geometric constraint method. First, the method proposed in the present invention has lower algorithm time consumption than the spatial geometric constraint. Secondly, relying on high-precision maps to provide road edge information, the scene adaptability of the algorithm of the present invention is improved. Finally, based on the road edge information provided by high-precision maps, the algorithm proposed in the present invention further solves the extraction of ground point clouds on large curvature slopes by designing an algorithm flow. On the whole, the laser radar ground point cloud extraction algorithm proposed in the present invention has a clear process, low time consumption, and good scene adaptability, while solving the difficulty of cleanly extracting ground point clouds on large curvature slopes.

[0080] like Figure 1 As shown, a method for quickly extracting laser radar ground point cloud includes the following steps:

[0081] S100. Acquire high-precision positioning information for characterizing the current position of the vehicle in real time; then convert the point cloud from the laser radar coordinate system to the vehicle body coordinate system based on the high-precision positioning information.

[0082] In this specific embodiment, in S100, the high-precision positioning module calculates, updates and sends the high-precision positioning information in real time.

[0083] The LiDAR software algorithm module receives high-precision positioning information in real time and converts the point cloud from the LiDAR coordinate system to the vehicle body coordinate system.

[0084] S200. Generate a road ROI area point set of the current vehicle's self-positioning information in real time; the road ROI area point set is located under the vehicle body coordinate system.

[0085] In this specific embodiment, after receiving the high-precision positioning information, S200 generates and dynamically sends the road ROI area point set where the current vehicle is located in real time according to the high-precision positioning information.

[0086] The points in the road ROI area point set include horizontal coordinates, vertical coordinates, and elevation coordinates.

[0087] The points in the road ROI area point set are taken from the static point set; the static point set is located under the vehicle body coordinate system, and is obtained and dynamically sent by the high-precision map based on the pre-built-in prior layer combined with the high-precision positioning information.

[0088] S300. Generate a contour area according to the road ROI area point set; and generate multiple small areas according to the road ROI area point set; the contour area is a collection of multiple small areas.

[0089] In this specific embodiment, generating a contour area according to the road ROI area point set in S300 specifically includes the following steps:

[0090] S310a. Receive the road ROI area point set sent by the high-precision map; then fit the road ROI area point set into a polygon closed area.

[0091] like Figure 2 As shown, the points in the polygon closed area only take the XY coordinates of each point, including the horizontal coordinate and the vertical coordinate; the polygon closed area is a two-dimensional polygonal plane area.

[0092] S320a. Filter the point cloud data generated by real-time scanning of the laser radar through the polygon closed area; the filtering rule only takes the XY coordinates of each point, filters according to the XY plane, and does not restrict z; then retain the point cloud inside the polygon closed area of ​​the polygon to obtain the contour area.

[0093] In this specific embodiment, in S300, multiple small areas are generated according to the road ROI area point set, which specifically includes the following steps:

[0094] S310b. Receive the road ROI area point set sent by the high-precision map.

[0095] S320b. Based on the road ROI area point set, multiple small areas are dynamically generated; the coordinates of the vertices of each small area are three-dimensional points, including abscissa, ordinate, and elevation coordinates.

[0096] It should be noted that S310b to S320b contain elevation information, which is different from S320a, which only divides the polygon area into multiple sub-areas according to the XY plane, but retains the polygon area information of the two-dimensional point set.

[0097] S400. Obtain a ground point cloud dataset and a non-ground point cloud dataset according to the contour area and the small area.

[0098] It should be noted that the principle of S400 is to loop through the multiple small areas generated by S320, loop through the point cloud inside the polygon after S310 filtering to determine whether it belongs to a certain small area, and then determine whether the point cloud in each small area belongs to the ground point in turn.

[0099] In this specific embodiment, S400 specifically includes the following steps:

[0100] S410. Receive the small areas obtained in S310b to S320b; then use the vertices of each small area to generate a corresponding three-dimensional cutting plane for each small area; the three-dimensional cutting plane is a surface structure in a three-dimensional coordinate system.

[0101] In this specific embodiment, generating a corresponding three-dimensional cutting plane for each small area in S410 specifically includes the following steps:

[0102] S411. Receive the road ROI area point set dynamically in real time.

[0103] S412. Divide the road ROI area point set into multiple small areas of different sizes.

[0104] It should be noted that the division in S412 is based on the XY plane, that is, the BEV perspective.

[0105] It should be further explained that the principle of S412 is: traverse and sort all the points sent by the high-precision map, and sort them from near to far (in the X-axis direction of the longitudinal distance from the vehicle). Connect the sorted point set into a closed sub-area in every 4 (2 points on the left and right edges). Gradually generate multiple sub-areas. If the remaining point set at the end of the traversal is less than 3, the last point is discarded. If it is greater than 3, they are connected to form a closed area.

[0106] In this specific embodiment, S412 divides the road ROI area point set into a plurality of small areas of different sizes, which specifically includes the following steps:

[0107] S412a traverses all points in the road ROI area point set sent by the high-precision map;

[0108] S412b. Sort all points in S412a in the X-axis direction according to the rule of near to far based on the longitudinal distance from the vehicle to form a sorted point set.

[0109] S412c. Connect the sorted point set into a closed sub-region in groups of 4; then interpret the generated closed sub-region as a small region.

[0110] In this specific embodiment, in S412c, every 4 connections form a closed sub-region, and the rule is: 2 points on each left and right edge.

[0111] S412d. According to the number of points remaining in the sorted point set, perform the following operations:

[0112] If the number of remaining points in the sorted point set is greater than 3, return and execute S412c again.

[0113] If the number of remaining points in the sorted point set is equal to 3, these 3 points are connected to form a closed subregion; then the generated closed subregion is interpreted as a small region.

[0114] If the number of remaining points in the sorted point set is less than 3, the last remaining point is discarded; then the process of S412 ends.

[0115] S413. Loop through each small area; solve the plane equation according to the vertex coordinates to obtain the three-dimensional tangent plane; the three-dimensional tangent plane is expressed as follows:

[0116] Ax+By+Cz+D=0

[0117] Among them: x is the horizontal coordinate; y is the vertical coordinate; z is the elevation coordinate; (A, B, C) is the normal vector; D is the distance from the origin to the plane.

[0118] It should be noted that each small area has at least 3 vertices.

[0119] In this specific embodiment, solving the plane equation in S413 specifically includes the following steps:

[0120] S413a. Obtain the number of vertices of the small area corresponding to the plane equation to be solved.

[0121] S413b. Perform the following operations based on the number of vertices in the small area:

[0122] If the number of vertices in the small area is greater than 3, the least squares method is used to solve the plane equation.

[0123] If the number of vertices in the small area is 3, the overdetermined equation solution is directly obtained through the vertex information to solve the plane equation.

[0124] S420. Loop through each three-dimensional cutting plane, and determine the corresponding three-dimensional cutting plane to which the point cloud in the polygon closed area belongs in turn; then divide the point cloud in the polygon closed area into each corresponding three-dimensional cutting plane.

[0125] S430. According to the distance from each point to the three-dimensional tangent plane, determine whether the point is above or below the three-dimensional tangent plane; then perform the following operations according to the determination result:

[0126] If the point is on the 3D tangent plane, the point is marked as a non-ground point.

[0127] If the point is below the 3D tangent plane, it is marked as a ground point.

[0128] S440. Pack all non-ground points to obtain a non-ground point cloud dataset in a polygon closed area; pack all ground points to obtain a ground point cloud dataset in a polygon closed area.

[0129] S500. Output ground point cloud dataset.

[0130] In order to further demonstrate the technical effect of the present invention, this specific embodiment also provides the following experimental data comparison, which can be intuitively compared with the effect of the prior art:

[0131] The experimental data comparison results are as follows Figure 3a , Figure 3b , Figure 4a , Figure 4b shown. Figure 4a This is the ground point extraction effect diagram of the traditional GPF algorithm, that is, the ground point extraction method introduced in the paper 1 involved in the background technology part. Figure 4b This is the effect of extracting ground points by the method of the present invention. It can be clearly seen that the ground points extracted by the present invention in the lower right corner are more accurate than the GPF proposed in Paper 1.

[0132] Figure 3a , Figure 3b That is, the point cloud effect of the ROI area after removing the ground points. Figure 3a The point cloud of the area of ​​interest after the traditional GPF algorithm extracts the ground points. Figure 3b is the point cloud of the region of interest of the method of the present invention. After comparison, it can be clearly seen that Figure 3a There are many noisy ground points that are not effectively filtered, which will lead to many false detections of point clouds in the area of ​​interest. Figure 3b A more accurate point cloud is Figure 3a The method does not generate a large number of false detections.

[0133] Attachment: Figure 3a and Figure 4a or Figure 3b and Figure 4b The experimental results of the example point cloud are merged together to form Figure 2 The point cloud of the polygon area.

[0134] It should be further explained that if you want to obtain a more accurate segmentation point cloud, the use of deep learning methods is better than this technical solution in terms of effect. At the same time, compared with this technical solution that relies on static point set information provided by high-precision maps, deep learning does not require high-precision map dependencies, but deep learning methods have higher data requirements and computing power costs. At present, deep learning segmentation solutions have not yet been widely used in the industry.

[0135] The key technical points of the present invention are summarized as follows:

[0136] 1. Generate polygonal contours for point cloud filtering using the three-dimensional point information of the road edge provided by the high-precision map.

[0137] 2. Generate multiple sub-regions for the polygonal contour and generate a three-dimensional tangent plane for each sub-region.

[0138] 3. Divide the point cloud into sub-region cutting planes to further refine and improve the ground point segmentation accuracy and precision of each sub-region, taking into account both precision and efficiency.

[0139] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0140] The disclosed embodiments are described above to enable any person skilled in the art to implement or use the present invention. Various modifications of these embodiments are obvious to those skilled in the art, and the general principles defined in the present invention may also be applied to other embodiments without departing from the spirit and scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments given in the present invention, but is consistent with the broadest scope of the principles and novel features disclosed in the present application.

[0141] The above description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described in the present invention are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

[0142] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for quickly extracting laser radar ground point clouds, characterized in that: The following steps are involved: S100. Acquire high-precision positioning information for characterizing the current position of the vehicle in real time; then convert the point cloud from the laser radar coordinate system to the vehicle body coordinate system according to the high-precision positioning information; S200. Generate in real time a road ROI area point set of the current vehicle's self-positioning information; the road ROI area point set is located under the vehicle body coordinate system; wherein: In S200, after receiving the high-precision positioning information, the high-precision map generates and dynamically sends the road ROI area point set where the current vehicle is located in real time according to the high-precision positioning information; S300. Generate a contour area according to the road ROI area point set; and generate multiple small areas according to the road ROI area point set; the contour area is a collection of multiple small areas; wherein: The step of generating a contour area according to the road ROI area point set in S300 specifically includes the following steps: S310a. Receive the road ROI area point set sent by the high-precision map; then fit the road ROI area point set into a polygon closed area; the points in the polygon closed area only take the XY coordinates of each point, including the horizontal coordinate and the vertical coordinate; the polygon closed area is a two-dimensional polygonal plane area; S320a. Through the polygon closed area, the point cloud data generated by the real-time scanning of the laser radar is filtered; the filtering rule only takes the XY coordinates of each point, filters according to the XY plane, and does not restrict z; then retains the point cloud inside the polygon closed area of ​​the polygon to obtain the contour area; The step of generating a plurality of small areas according to the road ROI area point set in S300 specifically includes the following steps: S310b receiving the road ROI area point set sent by the high-precision map; S320b. According to the road ROI area point set, a plurality of the small areas are dynamically generated; the coordinates of the vertices of each of the small areas are three-dimensional points, including abscissas, ordinates, and elevation coordinates; S400. Obtain a ground point cloud data set and a non-ground point cloud data set according to the contour area and the small area; wherein: S400 specifically includes the following steps: S410. Receive the small area obtained in S310b to S320b; then use the vertices of each small area to generate a corresponding three-dimensional cutting plane for each small area; the three-dimensional cutting plane is a surface structure in a three-dimensional coordinate system; S420. Loop through each of the three-dimensional cutting planes, and determine the corresponding three-dimensional cutting planes to which the point clouds in the polygon closed area belong in turn; then divide the point clouds in the polygon closed area into each of the corresponding three-dimensional cutting planes; S430. According to the distance from each point to the three-dimensional tangent plane, determine whether the point is above or below the three-dimensional tangent plane; then perform the following operations according to the determination result: If the point is above the three-dimensional tangent plane, the point is marked as a non-ground point; If the point is below the three-dimensional tangent plane, the point is marked as a ground point; S440. Pack all the non-ground points to obtain the non-ground point cloud data set in the polygon closed area; pack all the ground points to obtain the ground point cloud data set in the polygon closed area; S500. Output the ground point cloud dataset.

2. The method for rapidly extracting laser radar ground point clouds according to claim 1, characterized in that: In S100, the high-precision positioning module calculates, updates and sends the high-precision positioning information in real time; The laser radar software algorithm module receives the high-precision positioning information in real time and converts the point cloud from the laser radar coordinate system to the vehicle body coordinate system.

3. The method for rapidly extracting laser radar ground point clouds according to claim 2, characterized in that: The points in the road ROI area point set include horizontal coordinates, vertical coordinates, and elevation coordinates; the points in the road ROI area point set are taken from a static point set; the static point set is converted to the vehicle body coordinate system based on high-precision positioning information, and is obtained and dynamically sent by the high-precision map based on a pre-built-in prior layer combined with the high-precision positioning information.

4. The method for rapidly extracting laser radar ground point clouds according to claim 3, characterized in that: S410 generates a corresponding three-dimensional cutting plane for each small area, specifically including the following steps: S411. Real-time dynamic receiving of the road ROI area point set; S412. Dividing the road ROI area point set into a plurality of small areas of different sizes; S413. Loop through each of the small areas; solve the plane equation according to the vertex coordinates to obtain the three-dimensional tangent plane; the three-dimensional tangent plane is expressed as follows: Ax+By+Cz+D=0 Among them: x is the horizontal coordinate; y is the vertical coordinate; z is the elevation coordinate; (A, B, C) is the normal vector; D is the distance from the origin to the plane.

5. The method for rapidly extracting laser radar ground point clouds according to claim 4, characterized in that: Solving the plane equation in S413 specifically includes the following steps: S413a. Obtain the number of vertices of the small area corresponding to the plane equation to be solved; S413b. Perform the following operations according to the number of vertices in the small area: If the number of vertices in the small area is greater than 3, the least square method is used to solve the plane equation; If the number of vertices in the small area is 3, the solution of the overdetermined equation is directly obtained through the vertex information to solve the plane equation.

6. The method for rapidly extracting laser radar ground point clouds according to claim 5, characterized in that: The step of dividing the road ROI area point set into a plurality of small areas of different sizes in S412 specifically includes the following steps: S412a traverses all points in the road ROI area point set sent by the high-precision map; S412b. Sorting all points in S412a in the X-axis direction from the longitudinal distance of the vehicle according to the rule from near to far to form a sorted point set; S412c. The sorted point set is connected into a closed sub-region every 4 points; and the generated closed sub-region is interpreted as a small region; S412d. According to the number of points remaining in the sorted point set, perform the following operations: If the number of remaining points in the sorted point set is greater than 3, return and execute S412c again; If the number of remaining points in the sorted point set is equal to 3, then these 3 points are connected to form a closed sub-region; then the generated closed sub-region is interpreted as one of the small regions; If the number of remaining points in the sorted point set is less than 3, the last remaining point is discarded; then the process of S412 ends.

7. The method for rapidly extracting laser radar ground point clouds according to claim 6, characterized in that: As described in S412c, every 4 connections form a closed sub-region, and the rule is: 2 points on each of the left and right edges.

Images