Motor vehicle lane and sidewalk segmentation method based on vehicle-mounted point cloud

Abstract

The invention relates to the technical field of point cloud processing, in particular to a vehicle-mounted point cloud-based motorway and sidewalk segmentation method, which comprises the following steps of: acquiring road point cloud data; by calculating the normal vector and perpendicularity of the point clouds, the point clouds of which the perpendicularity is within a preset perpendicularity threshold range are reserved, and flat region point clouds are obtained; segmenting the point cloud of the flat region at equal time intervals by using a time index attribute; performing clustering blocking on each section of point cloud by utilizing hierarchical clustering, and reserving a clustering cluster point cloud with the maximum point number as a ground point cloud; dividing each section of ground point cloud into blocks, and obtaining boundary point clouds of a motor vehicle lane and a sidewalk by using a continuous bounded elevation judgment criterion; and segmenting the ground point cloud by using the boundary point cloud, and dividing sidewalks and motor vehicle lanes according to the point cloud number. According to the method, the spatial features and the gradient bounded features of the point cloud data are fully utilized, and efficient segmentation of the motor vehicle lane and the sidewalk is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud processing, and in particular to a method for segmenting a motor vehicle lane and a sidewalk of a vehicle-mounted point cloud. Background Art

[0002] Accurate segmentation of motor lanes and sidewalks is of great significance in urban traffic management, intelligent driving and road planning. Traditional methods for segmenting motor lanes and sidewalks usually rely on manual annotation or image-based detection technology.

[0003] The patent with publication number CN117745793A adopts a small sample instance segmentation model based on FGN to detect and divide the image to be measured, and generate multiple mask matrices, each mask matrix identifies a slow road area and area type; wherein the small sample instance segmentation model based on FGN is obtained by machine learning training using the COCO dataset and the SS dataset; this method is not only time-consuming and labor-intensive, but also requires the construction of a network model, and is easily affected by factors such as lighting conditions, weather changes and obstructions, resulting in insufficient accuracy and robustness of the segmentation results.

[0004] Research using vehicle-mounted point cloud data has been widely used in areas such as road marking extraction, obstacle detection, and building 3D modeling. However, there is still little research on the automatic segmentation of motor vehicle lanes and sidewalks, and most of the research focuses on segmentation methods for simple road scenes with curbs of obvious heights. These methods have difficulty segmenting situations where motor vehicle lanes and sidewalks are of similar or slowly changing heights, and have poor adaptability to complex road scenes. Summary of the invention

[0005] In view of the shortcomings of the existing methods, the present invention makes full use of the spatial characteristics and gradient bounded characteristics of point cloud data to achieve efficient segmentation of motor vehicle lanes and sidewalks.

[0006] The technical solution adopted by the present invention is: a method for segmenting motor vehicle lanes and sidewalks of a vehicle-mounted point cloud comprises the following steps:

[0007] Step 1: Collect road point cloud data;

[0008] Step 2: By calculating the normal vector and verticality of the point cloud, the point cloud whose verticality is within a preset verticality threshold range is retained to obtain the flat area point cloud;

[0009] As a preferred embodiment of the present invention, step 2 specifically includes:

[0010] Step 21, detect the pth point in the point cloud, if point p is not processed, initialize the verticality of point p;

[0011] Step 22: With point p as the center, obtain the point cloud within the spherical neighborhood radius of the point as the neighborhood point set P of the point p ;

[0012] Step 23: P p Perform PCA analysis to obtain the eigenvalues ​​and eigenvectors of the three coordinate axes in the spatial coordinate system; p The eigenvector corresponding to the smallest eigenvalue in is taken as the normal vector N of point p p ;

[0013] Step 24: Calculate the verticality of point p: V p =1-|Nzp|, |*| represents the absolute value, N p The z-axis normal vector of

[0014] Step 25, calculate the verticality of all points;

[0015] Step 26: retain points whose verticality range is within the preset verticality threshold as flat area point clouds.

[0016] Step 3: Use the time index attribute to segment the flat area point cloud into equal time intervals;

[0017] Step 4: Use hierarchical clustering to cluster each point cloud segment, and retain the cluster point cloud with the largest number of points as the ground point cloud;

[0018] As a preferred embodiment of the present invention, step 4 specifically includes:

[0019] Step 41, initializing a number of points in each point cloud segment into a number of independent clusters;

[0020] Step 42, calculate the distance between any two points and construct a distance matrix D for each segment of the point cloud;

[0021] As a preferred implementation of the present invention, the distance between any two points is calculated using the Euclidean distance.

[0022] Step 43: Compare the minimum distance between all clusters with the set distance threshold to merge the clusters;

[0023] As a preferred embodiment of the present invention, step 43 specifically includes:

[0024] When the minimum distance between all clusters is less than the set distance threshold, the two closest clusters are merged into a new cluster and the number of clusters is updated;

[0025] When the minimum distance between all clusters is no less than the set distance threshold, clustering stops.

[0026] Step 44, recalculate the distance between the new cluster and other clusters;

[0027] Step 45, stop until the minimum distance between all clusters is not less than the set distance threshold;

[0028] Step 46: retain the cluster with the largest number of points as the ground point cloud.

[0029] Step 5: Divide each segment of ground point cloud into blocks, and use the continuous bounded elevation judgment criterion to obtain the boundary point cloud of the motor vehicle lane and the sidewalk;

[0030] As a preferred implementation of the present invention, step five specifically includes:

[0031] Step 51, dividing the ground point cloud into blocks along the X axis;

[0032] As a preferred embodiment of the present invention, the blocks are divided according to a fixed grid size.

[0033] Step 52, calculate the number of points in each point cloud, and select the block with the most points as the central block;

[0034] Step 53, calculating the average elevation of the point cloud in each block as the representative value of each block;

[0035] Step 54, setting the center block and its left and right outer block values;

[0036] Step 55: Compare the center block with the left and right outer block values ​​to perform boundary point cloud recognition.

[0037] As a preferred embodiment of the present invention, step 55 specifically includes:

[0038] If the center block value V G0 The largest value indicates that the ground point cloud of this section is a motor vehicle lane, there is no sidewalk, and no processing is performed;

[0039] If the left outer block value V GLmax If the value is the largest, it means that there is a sidewalk in the left outer block and a motor vehicle lane in the right outer block, and the left outer block is processed continuously;

[0040] If the right outer block V GRmax If the value is the largest, it means that there is a sidewalk in the right outer block and the left outer block is a motor vehicle lane. Continue to process the right outer block; start from the center block and traverse the right outer blocks in sequence; determine whether the current block meets the following conditions:

[0041] V GRmax -V current >V T And V Gc-Rmax >V current

[0042] Among them, V Tis the preset height difference threshold; V current is the value of the current block; V GRmax is the maximum value of the right block of the center block; V Gc-Rmax is the value of the outer block of the current block;

[0043] When the conditions are met, the current block is a boundary block and processing stops; if not, continue to judge the outer blocks; if there is no block that meets the conditions, the ground point cloud of this section is all motor vehicle lane point cloud, and there is no sidewalk point cloud.

[0044] Step 6: Use the boundary point cloud to segment the ground point cloud, and divide the sidewalk and motor vehicle lane according to the number of point clouds;

[0045] As a preferred implementation of the present invention, step six specifically includes:

[0046] Step 61: if there is no boundary block in the ground point cloud, all point clouds are retained; if there is a boundary block in the ground point cloud, the mean value of the X-axis component of the point cloud in the boundary block is used as the segmentation threshold for segmentation;

[0047] Step 62: The number of point clouds of the motor vehicle lane is greater than the number of point clouds of the sidewalk;

[0048] Step 63: Merge the motor vehicle lane point clouds and sidewalk point clouds of all segments.

[0049] As a preferred embodiment of the present invention, a vehicle-mounted point cloud motor vehicle lane and sidewalk segmentation system includes: a memory for storing instructions executable by a processor; a processor for executing instructions to implement a vehicle-mounted point cloud motor vehicle lane and sidewalk segmentation method.

[0050] Beneficial effects of the present invention:

[0051] 1. The present invention firstly uses verticality to distinguish the curb area and the flat area; secondly, the driving area is determined according to the characteristics of the vehicle-borne point cloud that the closer the distance is, the greater the density is, and the non-road area is removed; finally, according to the continuous boundedness of the road cross-section direction, the boundary point cloud of the motor vehicle lane and the sidewalk is extracted quickly and accurately;

[0052] 2. The method of the present invention has strong interpretability, high feasibility and simple operation, which greatly improves the efficiency and accuracy of automatic segmentation of motor vehicle lanes and pedestrian areas;

[0053] 3. The method of the present invention is easy to operate and has strong adaptability. It significantly improves the accuracy and efficiency of segmentation and provides important technical support for intelligent traffic management and road three-dimensional modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the method for segmenting motor vehicle lanes and sidewalks of a vehicle-mounted point cloud of the present invention;

[0055] Figure 2 It is a schematic diagram of normal vectors of different regions of the present invention;

[0056] Figure 3 is a rendering of a road scene of the present invention;

[0057] Figure 4 It is a rendering process diagram of the present invention including only motor vehicle lanes and sidewalks;

[0058] Figure 5 It is an elevation distribution map of a segment of point cloud of the present invention;

[0059] Figure 6 It is a block schematic diagram of a section of ground point cloud of the present invention. DETAILED DESCRIPTION

[0060] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0061] like Figure 1 As shown, a method for segmenting motor vehicle lanes and sidewalks of a vehicle-mounted point cloud includes the following steps:

[0062] Step 1: Collect road point cloud data;

[0063] Road point cloud data can be obtained through vehicle-mounted laser radar. Road point cloud data can be obtained in real time through the vehicle-mounted laser radar installed on the vehicle;

[0064] You can also use laser radar to scan the road, or generate 3D point cloud data from multiple 2D images; you can also directly purchase an existing road point cloud database;

[0065] Step 2: Recognize the flat area point cloud through the normal vector and verticality of the point cloud;

[0066] like Figure 2 It is a schematic diagram of normal vectors in different areas. By calculating the normal vector and verticality of the point cloud, the point cloud with verticality within the preset verticality threshold is retained as the flat area point cloud;

[0067] Specifically include:

[0068] Step 21: Detect the pth point in the vehicle point cloud. If the pth point has not been processed, initialize the verticality of the pth point in the vehicle point cloud to Ip=-1;

[0069] Step 22: With the pth point as the center, obtain the vehicle point cloud within the spherical neighborhood radius r of the pth point as the neighborhood point set P of the pth point. p; The best radius parameter is selected through experiments as r0 = 0.2 meters;

[0070] Step 23: Neighborhood point set P of the pth point p Perform PCA principal component analysis to obtain the three eigenvalues ​​and eigenvectors corresponding to the X, Y, and Z axes in the spatial coordinate system; p The eigenvector corresponding to the smallest eigenvalue in is taken as the normal vector N of the pth point p (N xp ,N yp ,N zp );

[0071] In the spatial coordinate system, the Y axis is along the road direction, the X axis is perpendicular to the road direction, and the Z axis is perpendicular to the road.

[0072] Step 24: Verticality V of the pth point p is 1 and N zp The absolute value difference of V p =1-|N zp |, |*| represent absolute values;

[0073] Step 25, traverse all points in the point cloud, repeat steps 21 to 24, and obtain the verticality of all points;

[0074] Step 26: Keep the verticality range in [0,V T ] is a flat area point cloud, V T is the preset verticality threshold; in this embodiment, the optimal verticality threshold is V T0 =0.1.

[0075] Through step 2, objects with large verticality, such as buildings, fences, street lights, curbstones with obvious height differences, and cars, can be removed; Figure 3 (a), (b), and (c) are point clouds of road scenes (true color rendering), point clouds of road scenes (verticality rendering), and point clouds of flat areas (verticality rendering), respectively.

[0076] Step 3: Use the time index attribute to segment the flat area point cloud into equal time intervals;

[0077] The time interval of each point cloud is T0. In this example, the optimal time interval T0 = 0.5 seconds;

[0078] In the subsequent processing unit, the point cloud is segmented, and each segment of the point cloud is segmented and processed at the same time interval.

[0079] Step 4: Use hierarchical clustering based on Euclidean distance to cluster each point cloud segment, and retain the cluster point cloud with the largest number of points as the ground point cloud;

[0080] The specific steps are as follows:

[0081] Step 41, assuming that each segment of the point cloud has N points, each point is initialized into an independent cluster, so each segment has N clusters at the beginning;

[0082] Step 42: Use Euclidean distance as the metric to calculate the distance between any two points p i and p j The distance d(p i ,p j );

[0083] The Euclidean distance formula is

[0084] Each point cloud constructs a symmetric distance matrix D, where D ij Represents point p i and p j The Euclidean distance between

[0085] Step 43: Calculate whether the minimum distance between all clusters exceeds the set distance threshold D T , if the minimum distance is less than the distance threshold D T , merge the two closest clusters into a new cluster, which contains all the points of the original two clusters; after merging, update the number of clusters;

[0086] If the minimum distance between all clusters is greater than or equal to the distance threshold D T , then stop the clustering process;

[0087] In this embodiment, the optimal distance threshold D T =0.1 m;

[0088] Step 44, recalculate the distance between the new cluster and other clusters, the principle is the nearest point principle, that is, the distance between the new cluster and other clusters is the distance between the closest points of the two clusters;

[0089] Step 45: Repeat steps 43 and 44 until the minimum distance between all clusters exceeds the distance threshold D. T stop;

[0090] Step 46: retain the cluster with the largest number of points as the ground point cloud.

[0091] Through step 4, the point cloud after verticality screening and segmentation includes not only the road surface and the sidewalk, but also the curb plane in the center of the road, the flat area away from the road, the top of the street lamp, etc., so that the ground point cloud only includes the motor vehicle lane and the sidewalk.

[0092] Effects such as Figure 4(a), (b), and (c) show the segmented point cloud (verticality rendering), the segmented point cloud (domain point quantity rendering), and the clustered ground point cloud (true color rendering), respectively.

[0093] Step 5: Divide each segment of ground point cloud into blocks, and use the continuous bounded elevation judgment criterion to obtain the boundary point cloud of the motor vehicle lane and the sidewalk;

[0094] like Figure 5 The following is a schematic diagram of the block division of a certain ground point cloud. Step 5 specifically includes:

[0095] Step 51: Divide the ground point cloud into blocks along the X axis according to a fixed grid size GSD; the best grid size parameter in the embodiment is GSD=0.2 meters;

[0096] Step 52, calculate the number of points in each point cloud, and select the block with the most points as the central block;

[0097] Step 53: Calculate the average elevation of the point cloud in each block as the representative value V of each block. G ;

[0098] Step 54: Assume that the center block is G0, along the X axis, the block to the left of the center block is the left outer block G L , the block to the right of the center block is the right outer block G R The blocks with the largest values ​​on the left and right sides of the center block are respectively denoted as G Lmax , G Rmax ;

[0099] Step 55: record the block value of the center block as V G0 , the maximum block value of the left outer block is V GLmax , the maximum block value V of the right outer block GRmax ; Compare V G0 、V GLmax 、V GRmax The sizes of the three:

[0100] If V G0 If the value is the largest, it means that the ground point cloud in this section is a motor vehicle lane, there is no sidewalk, and no processing is done;

[0101] If V GLmax If the value is the largest, it means that there may be a sidewalk in the left outer block, and the right outer block is a motor vehicle lane, so continue to process the left outer block;

[0102] If V GRmax If the value is the largest, it means that there may be a sidewalk in the right outer block, and the left outer block is a motor vehicle lane, so continue to process the right outer block;

[0103] When V GRmaxThe largest, continue to process the right outer block; starting from the center block, traverse the right outer blocks in turn; each block processed is the current block G current , the block value of the current block is V current ; Cut-off G Rmax , the outer block of the current block is G c-Rmax , the corresponding block value is V Gc-Rmax ; Determine whether the current block meets the following conditions:

[0104] V GRmax -V current >V T And V Gc-Rmax >V current

[0105] Among them, V T To preset the height difference threshold;

[0106] If the current block meets the above conditions, the current block is a boundary block and processing stops; if not, continue to judge the outer blocks; if there is no block that meets the above conditions, the ground point cloud of this section is all motor vehicle lane point cloud, and there is no sidewalk point cloud.

[0107] Step 5 and Step 6 provide boundary point cloud.

[0108] Step 6: Use the boundary point cloud to segment the ground point cloud. The area with more points is the motorway point cloud, and the area with fewer points is the sidewalk point cloud.

[0109] The specific steps are as follows:

[0110] Step 61: If there is no boundary block in the ground point cloud, all point clouds are retained; if there is a boundary block in the ground point cloud, the mean value of the X component of the point cloud in the boundary block is used as the segmentation threshold to divide it into two parts;

[0111] Step 62: The portion with more points is the motorway point cloud, and the portion with fewer points is the sidewalk point cloud;

[0112] Step 63, merging the motorway point clouds and sidewalk point clouds of all segments;

[0113] like Figure 6 As shown, the boundary block and center block obtained through the point cloud, the dotted line on the left is the boundary block; the solid line on the right is the center block area; the left side of the green dotted line is the sidewalk, and the right side of the green dotted line is the motor vehicle lane.

[0114] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for segmenting motorway lanes and sidewalks of a vehicle-mounted point cloud, characterized in that: The following steps are involved: Step 1: Collect road point cloud data; Step 2: By calculating the normal vector and verticality of the point cloud, the point cloud whose verticality is within a preset verticality threshold range is retained to obtain the flat area point cloud; Step 3: Use the time index attribute to segment the flat area point cloud into equal time intervals; Step 4: Use hierarchical clustering to cluster each point cloud segment, and retain the cluster point cloud with the largest number of points as the ground point cloud; Step 5: Divide each segment of ground point cloud into blocks, and use the continuous bounded elevation judgment criterion to obtain the boundary point cloud of the motor vehicle lane and the sidewalk; Step 6: Use the boundary point cloud to segment the ground point cloud, and divide the sidewalk and motor vehicle lanes according to the number of point clouds.

2. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 1, characterized in that: Step 2 specifically includes: Step 21, detect the pth point in the point cloud, and if the point is not processed, initialize the verticality of the point; Step 22: With point p as the center, obtain the point cloud within the spherical neighborhood radius of the point as the neighborhood point set P of the point p ; Step 23: P p Perform PCA analysis to obtain the eigenvalues ​​and eigenvectors of the three coordinate axes in the spatial coordinate system; p The eigenvector corresponding to the smallest eigenvalue in is taken as the normal vector N of point p p ; Step 24: Calculate the verticality of point p: V p =1-|Nzp|, |*| represents the absolute value, N p The z-axis normal vector of Step 25, calculate the verticality of all points; Step 26: retain points whose verticality range is within the preset verticality threshold as flat area point clouds.

3. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 1, characterized in that: Step 4 specifically includes: Step 41, initializing a number of points in each point cloud segment into a number of independent clusters; Step 42, calculate the distance between any two points and construct a distance matrix D for each segment of the point cloud; Step 43: Compare the minimum distance between all clusters with the set distance threshold to merge the clusters; Step 44, recalculate the distance between the new cluster and other clusters; Step 45, stop until the minimum distance between all clusters is not less than the set distance threshold; Step 46: retain the cluster with the largest number of points as the ground point cloud.

4. The method for segmenting motor vehicle lanes and sidewalks of vehicle-mounted point clouds according to claim 3, characterized in that: Calculate the distance between any two points using Euclidean distance.

5. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 3, characterized in that: Step 43 specifically includes: When the minimum distance between all clusters is less than the set distance threshold, the two closest clusters are merged into a new cluster and the number of clusters is updated; When the minimum distance between all clusters is no less than the set distance threshold, clustering stops.

6. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 1, characterized in that: Step 5 specifically includes: Step 51, dividing the ground point cloud into blocks along the X axis; Step 52, calculate the number of points in each point cloud, and select the block with the most points as the central block; Step 53, calculating the average elevation of the point cloud in each block as the representative value of each block; Step 54, setting the center block and its left and right outer block values; Step 55: Compare the center block with the left and right outer block values ​​to perform boundary point cloud recognition.

7. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 6, characterized in that: The blocks are divided into fixed grid sizes.

8. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 6, characterized in that: Step 55 specifically includes: If the center block value V G0 The largest value indicates that the ground point cloud of this section is a motor vehicle lane, there is no sidewalk, and no processing is performed; If the left outer block value V GLmax If the value is the largest, it means that there is a sidewalk in the left outer block and a motor vehicle lane in the right outer block, and the left outer block is processed continuously; If the right outer block V GRmax If the value is the largest, it means that there is a sidewalk in the right outer block and the left outer block is a motor vehicle lane. Continue to process the right outer block; start from the center block and traverse the right outer blocks in sequence; determine whether the current block meets the following conditions: V GRmax -V current >V T and V Gc-Rmax >V current Among them, V T is the preset height difference threshold; V current is the value of the current block; V GRmax is the maximum value of the right block of the center block; V Gc-Rmax is the value of the outer block of the current block; When the conditions are met, the current block is a boundary block and processing stops; if not, continue to judge the outer blocks; if there is no block that meets the conditions, the ground point cloud of this section is all motor vehicle lane point cloud, and there is no sidewalk point cloud.

9. The method for segmenting motorway lanes and sidewalks of vehicle-mounted point clouds according to claim 1, characterized in that: Step 6 specifically includes: Step 61: if there is no boundary block in the ground point cloud, all point clouds are retained; if there is a boundary block in the ground point cloud, the mean value of the X-axis component of the point cloud in the boundary block is used as the segmentation threshold for segmentation; Step 62: The number of point clouds of the motor vehicle lane is greater than the number of point clouds of the sidewalk; Step 63: Merge the motor vehicle lane point clouds and sidewalk point clouds of all segments.

10. The vehicle-mounted point cloud segmentation system for motor vehicle lanes and sidewalks is characterized by: include: a memory for storing instructions executable by a processor; A processor, used to execute instructions to implement the method for segmenting motor vehicle lanes and sidewalks of a vehicle-mounted point cloud as described in any one of claims 1-9.

Images