Spatial correlation structured road complex scene point cloud classification segmentation method and system

Abstract

The invention discloses a spatial correlation structured road complex scene point cloud classification segmentation method and system, and relates to the technical field of road point cloud segmentation and classification, and the method comprises the steps: taking a laser radar as an original point, and constructing a polar coordinate grid map of a to-be-segmented laser point cloud; based on the statistical information of the point clouds in each grid, determining a grid type to which each grid belongs; performing feature extraction on grids on the polar coordinate grid map, and performing curve fitting to generate a boundary curve model; determining a current road direction based on the boundary curve model; associating grid types of a plurality of grids of the target grid along the current road direction, and dividing the target grid into a road inner obstacle grid and a road outer obstacle grid; and associating the grid types of the grids in the two side directions of the current road direction of the target grid, and carrying out type re-segmentation on the target grid. According to the invention, the technical problems of low boundary fitting accuracy and long time consumption in a complex scene in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of road point cloud segmentation and classification, and in particular to a spatially associated structured road complex scene point cloud classification segmentation method and system. Background Art

[0002] LiDAR is one of the core perception sensors of smart cars. In structured roads, the traversable area of ​​smart cars is relatively complex, but it is mainly the area within the sidewalk (i.e., road). Most of the point clouds obtained by single-frame scanning of LiDAR are located outside the road. This part of the point cloud not only interferes with the tracking and prediction of road obstacles by smart cars, but also reduces the timeliness of perception. Therefore, it needs to be filtered out in time. Road point clouds include ground point clouds and non-ground point clouds. The ground point cloud reflects the traversable area, and the non-ground point cloud (i.e., obstacle point cloud) reflects various dynamic and static obstacles. It is the data basis for perception tasks such as obstacle detection and target tracking, and is used in planning decision-making, tracking control and other links. The segmentation of road point clouds and the classification of obstacle ground attributes (referred to as classification) are basic tasks of the smart car environmental perception module, which are of great significance to improving the perception accuracy and efficiency of smart cars.

[0003] To achieve the segmentation and classification of road point clouds, relevant research contents include ground segmentation, road boundary extraction, point cloud segmentation and classification, etc. Among them, the problems to be solved in ground segmentation and road boundary extraction are complex and are the core research contents. The problems to be solved in segmentation and classification are relatively simple, and methods such as polygon point inclusion detection and three-point cosine calculation can be used. At present, through the preliminary research of experts and scholars, the ground point cloud segmentation effect is good and stable. Commonly used methods include methods based on raster maps, plane model fitting, and feature point extraction based on scan lines. Common methods for road boundary extraction include point cloud raster processing methods, cluster analysis methods, machine learning methods, and scan line methods. Among them, point cloud raster processing methods, cluster analysis methods, machine learning methods, etc. are not suitable for real-time road boundary extraction based on single-frame point clouds due to limitations in data density requirements, calculation amount, and preliminary sample requirements. The paper "Wang Guojun, Wu Jian, He Rui, et al. Speed ​​and Accuracy Tradeoff for LiDAR Data Based Road Boundary Detection [J]. IEEE / CAA Journal of Automatica Sinica" extracts road boundary feature points based on the height difference and smoothness of different point clouds on the same scan line of a single frame point cloud, and performs segmented processing in the front and back and left and right directions in the lidar coordinate system, and uses a polynomial curve to fit the road boundary. The method works well in straight road (curve) scenes, but the boundary fitting accuracy is low and very time-consuming in complex scenes such as intersections and multiple obstacles. The paper "ERNO H, CLAUDIU P, MIKLOS U. Real-Time LIDAR-Based Urban Road and Sidewalk Detection for Autonomous Vehicles [J] Sensors (Basel)" constructs the point cloud into two formats: scan line plus angle and rectangular segment. It uses X-zero, Z-zero and star search methods to obtain road boundary candidate points by windowing, and then selects boundary points from the candidate points angle by angle to generate a polygonal road boundary box. This method has a good overall road boundary segmentation effect, but the boundary feature candidate points have more noise, and there is a certain degree of missed segmentation and mis-segmentation. The time taken to process a single frame of this method is close to the time interval of the lidar scanning cycle. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for classifying and segmenting spatially associated structured road complex scene point clouds in order to solve at least one of the above technical problems.

[0005] In the first aspect, an embodiment of the present invention provides a spatially associated structured road complex scene point cloud classification and segmentation method, comprising: taking a laser radar as the origin, constructing a polar coordinate grid map of the laser point cloud to be segmented by a single frame scan of the laser radar; the polar coordinate grid map includes a plurality of grid blocks, each grid block includes a plurality of grids, and each grid includes a plurality of point clouds; based on the statistical information of the point cloud in each grid, determining the grid type to which each grid belongs; the grid types include: grid without point cloud, grid without plane obstacle, grid with plane obstacle, grid outside the road, and grid on the road ground; performing feature extraction on the grid on the polar coordinate grid map to obtain A characteristic grid is prepared, and curve fitting is performed on the characteristic grid to generate a boundary curve model; based on the boundary curve model, the current road direction is determined; the grid types of multiple grids of the target grid along the current road direction are associated, and the target grid is divided into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map; the grid types of the grids on both sides of the target grid along the current road direction are associated, and the type of the obstacle grid inside the road is re-segmented; the types after re-segmentation include an obstacle grid inside the road and a road ground grid.

[0006] Furthermore, the grid cell model of each grid includes:

[0007]

[0008] Among them, b i，j is the jth grid in the ith grid block, M is the number of grid blocks in the polar coordinate grid map, N is the number of grids in each grid block, θ min ,θ max 、r min 、r max They are the grid's starting angle, ending angle, path length starting value, path length ending value, n, z min 、z max 、z mea , σ are the number of point clouds in the grid, the maximum z-axis coordinate value of the point cloud, the minimum z-axis coordinate value of the point cloud, the average z-axis coordinate value of the point cloud, and the standard deviation of the z-axis coordinate value of the point cloud, respectively, and φ is the grid type.

[0009] Further, based on the statistical information of the point cloud in each grid, the grid type to which each grid belongs is determined, including: judging whether the number of point clouds in the grid to be determined is less than a preset number threshold; if so, determining the grid to be determined as a grid without point clouds; judging whether the height difference of the point cloud z-axis coordinates and the standard deviation of the point cloud z-axis coordinate values ​​in the grid to be determined are less than a preset height difference threshold and a preset standard deviation threshold, respectively; if so, determining the grid to be determined as a plane grid; if not, determining the grid to be determined as a non-plane obstacle grid; based on preset determination conditions, determining whether the grid corresponding to the plane grid is a road ground grid, a ground grid outside the road, or a plane obstacle grid; wherein the preset determination conditions include:

[0010]

[0011] Among them, η1 and η2 are the preset road ground height threshold and the preset road outside ground height threshold, respectively. and They are respectively the average height threshold of the preset road ground grid and the preset outer road ground height threshold, φ=3 indicates that the grid type is the plane obstacle grid, φ=4 indicates that the grid type is the outer road ground grid, and φ=5 indicates that the grid type is the road ground grid.

[0012] Furthermore, the extraction conditions of the feature grid include:

[0013]

[0014] in, is the acquisition length; performing curve fitting on the characteristic grid, including: using the RANSCA method and a quadratic polynomial to perform curve fitting on the point cloud in the characteristic grid.

[0015] Furthermore, based on the boundary curve model, the current road direction is determined, including: searching for the point on the right boundary in the boundary curve model that is closest to the target point on the left boundary and forming a horizontal vector with the target point; and taking a vector perpendicular to the horizontal vector as the current road direction.

[0016] Furthermore, associating the grid types of a plurality of grids of the target grid along the current road direction, and segmenting the target grid into an on-road obstacle grid and an off-road obstacle grid, comprises: counting the number of grid types of the plurality of grids of the target grid along the current road direction, which are road ground grids, off-road ground grids, and grids without point clouds, which are respectively a first statistical number, a second statistical number, and a third statistical number; determining whether the first statistical number is greater than a first preset threshold; if so, segmenting the target grid into an on-road obstacle grid; if not, determining whether the second statistical number is greater than a second preset threshold; if so, segmenting the target grid into an off-road obstacle grid.

[0017] Furthermore, the grid types of the grids on both sides of the target grid along the current road direction are associated, and the types of the obstacle grids on the road are re-segmented, including: determining whether the grid types of the grids on both sides of the target grid along the current road direction are both off-road ground grids; if so, re-segmenting the obstacle grids on the road into road ground grids.

[0018] In a second aspect, an embodiment of the present invention further provides a spatially associated structured road complex scene point cloud classification and segmentation system, comprising: a construction module, a classification module, an extraction module, a determination module, a segmentation module and a re-segmentation module; wherein the construction module is used to construct a polar coordinate grid map of the laser point cloud to be segmented by a single frame scan of the laser radar with the laser radar as the origin; the polar coordinate grid map includes a plurality of grid blocks, each grid block includes a plurality of grids, and each grid includes a plurality of point clouds; the classification module is used to determine the grid type to which each grid belongs based on the statistical information of the point cloud in each grid; the grid types include: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid outside the road, and grid on the road ground; the extraction module is used to classify the polar coordinate grid. The grid on the grid map is subjected to feature extraction to obtain a feature grid, and the feature grid is subjected to curve fitting to generate a boundary curve model; the determination module is used to determine the current road direction based on the boundary curve model; the segmentation module is used to associate the grid types of multiple grids of the target grid along the current road direction, and segment the target grid into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map; the re-segmentation module is used to associate the grid types of the grids on both sides of the target grid along the current road direction, and re-segment the type of the obstacle grid inside the road; the types after re-segmentation include an obstacle grid inside the road and a road ground grid.

[0019] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the computer program.

[0020] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method described in the first aspect is implemented.

[0021] The present invention provides a spatially associated structured road complex scene point cloud classification and segmentation method and system. On the basis of the point cloud polar coordinate grid map, the road ground grid is segmented by smoothness and height constraints, and the road obstacle grid is segmented from all obstacle grids based on the spatial correlation characteristics of the obstacle and the ground, thereby completing the classification and segmentation of the road point cloud, alleviating the technical problems of low boundary fitting accuracy, more noise in boundary feature candidate points and long time consumption in the prior art in complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A flowchart of a spatially associated structured road complex scene point cloud classification and segmentation method provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of constructing a polar coordinate grid map provided by an embodiment of the present invention;

[0025] Figure 3 A grid classification schematic diagram provided for an embodiment of the present invention;

[0026] Figure 4 A schematic diagram of point cloud data provided for an embodiment of the invention;

[0027] Figure 5 A grid type segmentation flow chart provided by an embodiment of the present invention;

[0028] Figure 6 A grid type re-segmentation flow chart provided in an embodiment of the present invention;

[0029] Figure 7A spatially associated structured road complex scene point cloud classification and segmentation system is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Embodiment 1

[0032] Figure 1 1 is a flow chart of a method for classifying and segmenting a spatially associated structured road complex scene point cloud according to an embodiment of the present invention. Figure 1 As shown, the method specifically comprises the following steps:

[0033] Step S102, taking the laser radar as the origin, constructing a polar coordinate grid map of the laser point cloud to be segmented by a single frame scan of the laser radar; wherein the polar coordinate grid map includes multiple grid blocks, each grid block includes multiple grids, and each grid includes multiple point clouds.

[0034] Step S104, based on the statistical information of the point cloud in each grid, determine the grid type to which each grid belongs, wherein the grid types include: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid on the ground outside the road, and grid on the ground of the road.

[0035] Step S106, extracting features from the grids on the polar coordinate grid map to obtain feature grids, and performing curve fitting on the feature grids to generate a boundary curve model.

[0036] Step S108, determining the current road direction based on the boundary curve model.

[0037] Step S110, associating the grid types of multiple grids of the target grid along the current road direction, dividing the target grid into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map.

[0038] Step S112, associating the grid types of the grids on both sides of the target grid along the current road direction, and re-segmenting the road obstacle grids into types; the types after re-segmentation include road obstacle grids and road ground grids.

[0039] Specifically, the number of point clouds obtained by a single-frame scan of the laser radar can reach tens of thousands, and dimensionality reduction processing is required to improve efficiency. Rasterization is an effective dimensionality reduction processing method, and the point cloud grid is usually constructed based on the polar coordinate system and the Cartesian coordinate system. In terms of single grid size, the polar coordinate grid exhibits the characteristics of "small near and large far", while the Cartesian coordinate grid is usually fixed. The hardware structure of the laser radar causes the point cloud distribution to appear "dense near and sparse far". In order to ensure the uniformity and quantity of the grid unit point cloud distribution, the present invention constructs a polar coordinate grid map with the laser radar as the origin, and the construction method is as follows: Figure 2 As shown, Figure 2 It is a schematic diagram of constructing a polar coordinate grid map provided according to an embodiment of the present invention.

[0040] Specifically, Figure 2 As shown, all grids b corresponding to the angle θ are defined to form a grid block S. The relationship between the grid angle α, grid path length r, grid map radius R, the number of grid blocks M, and the number of grids N contained in the grid block is as follows:

[0041]

[0042] The grid cell model for each grid consists of:

[0043]

[0044] Among them, b i,j is the jth grid in the ith grid block, M is the number of grid blocks in the polar coordinate grid map, N is the number of grids in each grid block, θ min ,θ max 、r min 、r max They are the grid's starting angle, ending angle, path length starting value, path length ending value, n, z min 、z max 、z mea , σ are the number of point clouds in the grid, the maximum z-axis coordinate value of the point cloud, the minimum z-axis coordinate value of the point cloud, the average z-axis coordinate value of the point cloud, and the standard deviation of the z-axis coordinate value of the point cloud, respectively, and φ is the grid type.

[0045] For any point cloud p(x, y, z) on the polar coordinate grid map, calculate its angle and path length according to the following formula to determine which grid b it belongs to. i，j , and update the corresponding grid statistics:

[0046]

[0047] In structured roads, the real environment around the smart car can be summarized into four types: road surface, road obstacles, sidewalks, and off-road obstacles; from the perspective of fitting processing, the grid can be divided into grids without point clouds, plane grids, and non-plane grids. Combining the real road environment and fitting processing, the overall grid classification is as follows: Figure 3 As shown, when φ takes a value of 1 to 5, it represents five types: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid with ground outside the road, and grid with ground on the road. Specifically, step S104 further includes the following steps:

[0048] Step S1041 , determining whether the number of point clouds in the grid to be determined is less than a preset number threshold; if so, determining the grid to be determined as a grid without point clouds.

[0049] Specifically, for judging whether there is a point cloud grid, the minimum point cloud quantity threshold is set as the preset quantity threshold. If the point cloud quantity in the grid is less than the preset quantity threshold, it is determined that there is no point cloud grid, otherwise it is a point cloud grid.

[0050] Step S1042, determine whether the point cloud z-axis coordinate height difference and the point cloud z-axis coordinate value standard deviation in the grid to be determined are respectively less than the preset height difference threshold and the preset standard deviation threshold; if so, determine the grid to be determined as a plane grid; if not, determine the grid to be determined as a non-planar obstacle grid.

[0051] Specifically, to determine whether it is a plane grid, first set the z-axis height difference threshold in the grid to the preset height difference threshold ε, and the z-axis coordinate value standard deviation threshold to the preset standard deviation threshold δ; when the following judgment formula is met, it is determined to be a plane grid:

[0052]

[0053] Step S1043, based on preset determination conditions, determining whether the grid corresponding to the plane grid is a road ground grid, a road outer ground grid or a plane obstacle grid; wherein the preset determination conditions include:

[0054]

[0055] Among them, η1 and η2 are the preset road ground height threshold and the preset road outside ground height threshold, respectively. and They are respectively the average height threshold of the preset road ground grid and the preset height threshold of the ground outside the road. φ=3 means the grid type is a plane obstacle grid, φ=4 means the grid type is a ground grid outside the road, and φ=5 means the grid type is a road ground grid. Specifically, φ=1 means a grid without point cloud, and φ=2 means a non-planar obstacle grid.

[0056] Figure 4is a schematic diagram of point cloud data provided according to an embodiment of the present invention. Figure 4 As shown in the figure, although the obstacle is located in the road, the grid where it is located is not judged as a ground grid due to the obstacle occlusion. However, the road is continuous. Along the road direction, the front side, the back side, and the side near the center of the road of the obstacle are all road ground, and the side away from the center of the road is the road ground or sidewalk. This is the spatial correlation characteristic between the obstacle and the ground.

[0057] The premise for using spatial correlation characteristics to achieve road obstacle point cloud segmentation is that the road direction is known. Affected by factors such as the heading of the smart car and the shape of the road, the positive direction of the coordinate system does not necessarily coincide with the road direction. The direction of the road cannot be determined by the direction of the coordinate system. It is necessary to use the grid as the basic unit to obtain the approximate road direction. Specifically, the embodiment of the present invention uses the grid as the basic unit, extracts the feature grid, roughly fits the road boundary, and then obtains the approximate direction of the road.

[0058] When extracting feature grids, for grid bi,j, in the grid block Si to which it belongs, if the following extraction conditions of feature grids are met, the grid is judged to be a feature grid:

[0059]

[0060] in, is the acquisition length.

[0061] In an optional implementation provided by an embodiment of the present invention, performing curve fitting on the feature grid in step S106 includes: performing curve fitting on the point cloud in the feature grid using a RANSCA method and a quadratic polynomial.

[0062] Specifically, step S108 also includes the following steps:

[0063] Step S1081, searching for a point on the right boundary of the boundary curve model that is closest to the target point on the left boundary, and forming a horizontal vector with the target point;

[0064] Step S1082, taking the perpendicular vector to the horizontal vector as the current road direction.

[0065] Specifically, the present invention constructs a polar coordinate grid map with the laser radar as the origin. The construction method is to use the laser radar as the coordinate center, the front of the laser radar is the positive direction of the X-axis, the right of the laser radar is the positive direction of the Y-axis, and the top of the laser radar is the positive direction of the Z-axis. On the left and right sides of the laser radar, there are road edges, that is, actual curbs, which are the left and right boundaries. After obtaining the feature grid, the present invention is divided into left and right boundaries. The RANSCA method is used to roughly fit a quadratic polynomial curve to each boundary. After the fitting is completed, outliers with errors greater than δ are eliminated. In order to ensure that the number of curve points is too small, equidistant interpolation is performed according to the curve model. Then, the points closest to the left boundary are searched on the right boundary in turn to form a vector β, and its perpendicular vector β ⊥ This is the road direction of the current location.

[0066] The non-planar obstacle grid and the planar obstacle grid obtained by the above judgment may be the obstacle grid in the road. Based on the segmentation of the road ground grid, the present invention segments the road obstacle grid based on the spatial correlation characteristics between the obstacle and the ground.

[0067] Specifically, step S110 includes:

[0068] The grid types of the multiple grids of the target grid along the current road direction are road ground grids, off-road ground grids, and grids without point clouds, which are respectively the first statistical quantity, the second statistical quantity, and the third statistical quantity;

[0069] Determine whether the first statistical quantity is greater than a first preset threshold; if so, segment the target grid into road obstacle grids;

[0070] If not, it is determined whether the second statistical quantity is greater than a second preset threshold; if yes, the target grid is segmented into off-road obstacle grids.

[0071] Figure 5 is a grid type segmentation flow chart provided according to an embodiment of the present invention. Figure 5 As shown, for the single-direction information judgment process, C plane grids are selected along the road direction during judgment, and the C grids are classified and counted. The number of road ground grids, road ground grids, and grids without point clouds are the first statistical number C1, the second statistical number C2, and the third statistical number C3, respectively. Figure 5 The “quantity satisfies” in the above sentence refers to whether the number of each type of grids counted meets the preset threshold. For example, the number thresholds of road ground grids, off-road ground grids, and grids without point clouds are set as the first threshold C 1 , the second threshold C 2 , the third threshold C 3 The judgment process is to first determine whether the number of ground grids counted is greater than C 1, if it is greater than, the current grid is judged to belong to the road ground grid (i.e. Figure 5 If it does not meet the requirement, count the number of ground grids outside the road. If it is greater than C 2 , then the grid is the ground grid outside the road (i.e. Figure 5 If none of them meet the requirements, the current grid type cannot be determined (i.e. Figure 5 cannot be determined).

[0072] Specifically, step S112 includes: determining whether the grid types of the grids on both sides of the target grid along the current road direction are both off-road ground grids; if yes, resegmenting the on-road obstacle grids into road ground grids.

[0073] Figure 6 is a grid type re-segmentation flow chart provided according to an embodiment of the present invention. Figure 6 As shown in the figure, for the process of multi-directional information segmentation of road obstacle grid, according to Figure 6 The process shown in the figure is to re-segment the grid type of the obstacle grid in the road. Figure 6 As shown, first determine whether both sides are roads. If yes, or one side is a road, then determine it as an obstacle grid within the road (i.e. Figure 6 If there is no road on both sides, the road barrier grid is re-segmented into a road ground grid (i.e. Figure 6 If the two sides cannot be determined, the center side is judged to be a road; if the center side is not a road, the road barrier grid is re-segmented into a road ground grid; if the center side is a road, the boundary side is judged to be a road, if yes, it is determined to be a road barrier grid, if not, the road barrier grid is re-segmented into a road ground grid.

[0074] From the above description, it can be seen that the present invention provides a spatially associated structured road complex scene point cloud classification and segmentation method. On the basis of the point cloud polar coordinate grid map, the road ground grid is segmented by smoothness and height constraints, and the road obstacle grid is segmented from all obstacle grids based on the spatial correlation characteristics of the obstacle and the ground, thereby completing the classification and segmentation of the road point cloud, alleviating the technical problems of low boundary fitting accuracy, high noise of boundary feature candidate points and long time consumption in the prior art in complex scenes.

[0075] Embodiment 2

[0076] Figure 7 A spatially associated structured road complex scene point cloud classification and segmentation system is provided according to an embodiment of the present invention, characterized in that it includes: a construction module 10, a classification module 20, an extraction module 30, a determination module 40, a segmentation module 50 and a re-segmentation module 60.

[0077] Specifically, the construction module 10 is used to construct a polar coordinate grid map of the laser point cloud to be segmented by the laser radar single frame scan with the laser radar as the origin; the polar coordinate grid map includes a plurality of grid blocks, each grid block includes a plurality of grids, and each grid includes a plurality of point clouds;

[0078] The classification module 20 is used to determine the grid type to which each grid belongs based on the statistical information of the point cloud in each grid; the grid types include: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid on the ground outside the road, and grid on the ground of the road;

[0079] The extraction module 30 is used to extract features from the grids on the polar coordinate grid map to obtain feature grids, and to perform curve fitting on the feature grids to generate a boundary curve model;

[0080] A determination module 40, for determining a current road direction based on a boundary curve model;

[0081] The segmentation module 50 is used to associate the grid types of multiple grids of the target grid along the current road direction, and segment the target grid into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map;

[0082] The re-segmentation module 60 is used to associate the grid types of the grids on both sides of the target grid along the current road direction, and re-segment the road obstacle grid by type; the types after re-segmentation include road obstacle grid and road ground grid.

[0083] Specifically, the classification module 20 is further used for:

[0084] Determine whether the number of point clouds in the grid to be determined is less than a preset number threshold; if so, determine the grid to be determined as a grid without point clouds;

[0085] Determine whether the point cloud z-axis coordinate height difference and the point cloud z-axis coordinate value standard deviation in the grid to be determined are less than a preset height difference threshold and a preset standard deviation threshold, respectively; if yes, determine the grid to be determined as a plane grid; if not, determine the grid to be determined as a non-planar obstacle grid;

[0086] Based on the preset determination conditions, it is determined whether the grid corresponding to the plane grid is a road ground grid, an off-road ground grid or a plane obstacle grid.

[0087] Specifically, the determination module 40 is further used for:

[0088] Find the point on the right boundary in the boundary curve model that is closest to the target point on the left boundary and form a horizontal vector with the target point;

[0089] The perpendicular vector to the horizontal vector is taken as the current road direction.

[0090] Specifically, the segmentation module 50 is further used for:

[0091] The grid types of the multiple grids of the target grid along the current road direction are road ground grids, off-road ground grids, and grids without point clouds, which are respectively the first statistical quantity, the second statistical quantity, and the third statistical quantity;

[0092] Determine whether the first statistical quantity is greater than a first preset threshold; if so, segment the target grid into road obstacle grids;

[0093] If not, it is determined whether the second statistical quantity is greater than a second preset threshold; if yes, the target grid is segmented into off-road obstacle grids.

[0094] The re-segmentation module 60 is further used to: determine whether the grid types of the grids on both sides of the target grid along the current road direction are both off-road ground grids; if yes, re-segment the on-road obstacle grid into road ground grids.

[0095] The present invention further provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method provided in the embodiment of the present invention when executing the computer program.

[0096] The present invention also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the method provided in the embodiment of the present invention is implemented.

[0097] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0098] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A spatially associated structured road complex scene point cloud classification and segmentation method, characterized in that: include: Taking the laser radar as the origin, constructing a polar coordinate grid map of the laser point cloud to be segmented by the single-frame scan of the laser radar; the polar coordinate grid map includes a plurality of grid blocks, each grid block includes a plurality of grids, and each grid includes a plurality of point clouds; Based on the statistical information of the point cloud in each grid, determine the grid type to which each grid belongs; the grid types include: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid on the ground outside the road, and grid on the ground of the road; Extracting features from the grids on the polar coordinate grid map to obtain feature grids, and performing curve fitting on the feature grids to generate a boundary curve model; Determining a current road direction based on the boundary curve model; Associating the grid types of multiple grids of the target grid along the current road direction, dividing the target grid into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map; The grid types of the grids on both sides of the target grid along the current road direction are associated, and the road obstacle grid is re-segmented into types; the types after re-segmentation include road obstacle grids and road ground grids.

2. The method according to claim 1, characterized in that: The grid cell model for each grid consists of: Among them, b i,j is the jth grid in the ith grid block, M is the number of grid blocks in the polar coordinate grid map, N is the number of grids in each grid block, θ min ,θ max 、r min 、r max They are the grid's starting angle, ending angle, path length starting value, path length ending value, n, z min 、z max 、z mea , σ are the number of point clouds in the grid, the maximum z-axis coordinate value of the point cloud, the minimum z-axis coordinate value of the point cloud, the average z-axis coordinate value of the point cloud, and the standard deviation of the z-axis coordinate value of the point cloud, respectively, and φ is the grid type.

3. The method according to claim 2, characterized in that: Based on the statistical information of the point cloud within each grid, the grid type of each grid is determined, including: Determine whether the number of point clouds in the grid to be determined is less than a preset number threshold; if so, determine the grid to be determined as a grid without point clouds; Determine whether the point cloud z-axis coordinate height difference and the point cloud z-axis coordinate value standard deviation in the grid to be determined are respectively less than a preset height difference threshold and a preset standard deviation threshold; if yes, determine the grid to be determined as a plane grid; if not, determine the grid to be determined as a non-planar obstacle grid; Based on preset determination conditions, determine whether the grid corresponding to the plane grid is a road ground grid, a road outside ground grid or a plane obstacle grid; wherein the preset determination conditions include: Among them, η1 and η2 are the preset road ground height threshold and the preset road outside ground height threshold, respectively. and They are respectively the average height threshold of the preset road ground grid and the preset outer road ground height threshold, φ=3 indicates that the grid type is the plane obstacle grid, φ=4 indicates that the grid type is the outer road ground grid, and φ=5 indicates that the grid type is the road ground grid.

4. The method according to claim 3, characterized in that: The extraction conditions of the feature grid include: in, is the acquisition length; Performing curve fitting on the feature grid includes: performing curve fitting on the point cloud in the feature grid using a RANSCA method and a quadratic polynomial.

5. The method according to claim 1, characterized in that: Determining a current road direction based on the boundary curve model includes: Finding a point on the right boundary of the boundary curve model that is closest to a target point on the left boundary, and forming a horizontal vector with the target point; The perpendicular vector to the horizontal vector is taken as the current road direction.

6. The method according to claim 1, characterized in that: Associating the target grid with the grid types of multiple grids along the current road direction, and dividing the target grid into an obstacle grid inside the road and an obstacle grid outside the road, including: Counting the number of grid types of a plurality of grids of the target grid along the current road direction, which are road ground grids, off-road ground grids, and grids without point clouds, which are first counted number, second counted number, and third counted number, respectively; Determine whether the first statistical quantity is greater than a first preset threshold; if so, segment the target grid into road obstacle grids; If not, it is determined whether the second statistical quantity is greater than a second preset threshold; if yes, the target grid is segmented into off-road obstacle grids.

7. The method according to claim 1, characterized in that: Associating the grid types of the grids on both sides of the target grid along the current road direction, and performing type re-segmentation on the obstacle grids in the road, including: It is determined whether the grid types of the grids on both sides of the target grid along the current road direction are both off-road ground grids; if yes, the on-road obstacle grid is re-segmented into road ground grids.

8. A spatially associated structured road complex scene point cloud classification and segmentation system, characterized in that: include: Construction module, classification module, extraction module, determination module, segmentation module and re-segmentation module; among them, The construction module is used to construct a polar coordinate grid map of the laser point cloud to be segmented by the laser radar single frame scan with the laser radar as the origin; the polar coordinate grid map includes a plurality of grid blocks, each grid block includes a plurality of grids, and each grid includes a plurality of point clouds; The classification module is used to determine the grid type to which each grid belongs based on the statistical information of the point cloud in each grid; the grid types include: grid without point cloud, grid with non-planar obstacle, grid with planar obstacle, grid on the ground outside the road, and grid on the ground of the road; The extraction module is used to extract features from the grids on the polar coordinate grid map to obtain feature grids, and to perform curve fitting on the feature grids to generate a boundary curve model; The determination module is used to determine the current road direction based on the boundary curve model; The segmentation module is used to associate the grid types of multiple grids of the target grid along the current road direction, and segment the target grid into an obstacle grid inside the road and an obstacle grid outside the road; the target grid is a grid whose grid type is a non-planar obstacle grid or a planar obstacle grid in the polar coordinate grid map; The re-segmentation module is used to associate the grid types of the grids on both sides of the target grid along the current road direction, and re-segment the types of the road obstacle grids; the types after re-segmentation include road obstacle grids and road ground grids.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images