A lane detection method based on lidar point cloud information

Abstract

The invention discloses a lane line detection method based on laser radar point cloud information, the method comprising: obtaining vehicle laser radar point cloud information, extracting point cloud of area near the ground from the point cloud, and extracting the ground from it; performing multi-frame accumulation on ground points; generating an intensity distribution histogram according to the ground points accumulated in multiple frames, obtaining an intensity threshold for extracting lane line point cloud based on the intensity distribution histogram, and determining lane line candidate points; using histogram filtering to remove noise points in lane line candidate points; using a DBSCAN algorithm to cluster multiple point cloud points close in position to generate multiple point cloud clusters; using a union-find algorithm to cluster point cloud clusters located on the same lane line into one category; using a least squares fitting method to fit the lane line points in each category, obtaining a lane line equation, and realizing lane line detection; the lane line detection method has high real-time performance and high precision, effectively filters out noise points, and reduces missed detection and false detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of lane line detection, and in particular relates to a lane line detection method based on laser radar point cloud information. Background Art

[0002] In today's rapidly developing intelligent driving, in order for an autonomous vehicle to have the ability to drive intelligently on the road, it needs to have basic lane line recognition capabilities, be able to perceive nearby lane lines, and obtain a variety of lane line location information, which can also be used for subsequent positioning, decision-making, and planning modules. Due to the color attributes of lane lines, lane line detection has been widely studied in the field of image recognition, but images are easily affected by lighting. With the development of lidar technology, lane line detection technology using laser point cloud information has also attracted the attention of many scholars, but there are still many problems such as low lane line recall rate, many noise points, and false detection; and the currently commonly used deep learning methods require the support of labeled data sets and a long training process, with poor real-time performance and low detection efficiency and accuracy.

[0003] Therefore, how to meet the real-time requirements in lane line detection, effectively filter out noise points, reduce missed detections and false detections, and improve lane line detection accuracy has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0004] In view of the above problems, the present invention provides a lane line detection method based on laser radar point cloud information that at least solves some of the above technical problems. The method has high real-time performance and high precision in lane line detection, effectively filters out noise points, and reduces missed detections and false detections.

[0005] An embodiment of the present invention provides a lane line detection method based on laser radar point cloud information, the method at least comprising:

[0006] S1. Obtaining laser radar point cloud information of the vehicle, and extracting point cloud of the area near the ground from the point cloud;

[0007] S2, extracting a ground area from the point cloud of the ground vicinity as a target area;

[0008] S3, performing multi-frame accumulation of ground points in the target area;

[0009] S4, generating an intensity distribution histogram according to the ground points accumulated in multiple frames, obtaining an intensity threshold for extracting the lane line point cloud based on the intensity distribution histogram, and determining the lane line candidate points according to the intensity threshold;

[0010] S5. Using a histogram filtering method to remove noise points from the lane line candidate points;

[0011] S6. Use the DBSCAN algorithm to cluster multiple point cloud points that are close in position to generate multiple point cloud clusters;

[0012] S7, clustering the multiple point cloud clusters using a union-find algorithm, and clustering the point cloud clusters located on the same lane line into one category;

[0013] S8. Use the least squares fitting method to fit each type of lane line point to obtain the lane line equation, and implement lane line detection based on the lane line equation.

[0014] Furthermore, the step S1 specifically includes: acquiring laser radar point cloud information of the vehicle, and using the cuboid filter in the PCL library to extract the point cloud within the cuboid area near the ground from the point cloud information.

[0015] Furthermore, in step S3, multi-frame accumulation of ground points in the target area includes:

[0016] By using the change in GPS data between two frames of the ground point, the point cloud coordinates are transformed, and the multi-frame point clouds near the current frame are converted to the current frame coordinate system, and the multi-frames complement each other.

[0017] Furthermore, in step S4, obtaining an intensity threshold for extracting a lane line point cloud based on the intensity distribution histogram, and determining lane line candidate points according to the intensity threshold, includes:

[0018] Traversing the intensity of the point cloud from small to large on the intensity distribution histogram, setting filtering conditions for the number of points, gradient, and intensity respectively, and finding the first intensity value that meets the three conditions that the number of points is less than the point number threshold, the gradient is less than the gradient threshold, and the intensity is greater than the intensity threshold, as the separation threshold to be obtained;

[0019] For each frame, the separation threshold is repeatedly obtained, and the point cloud with intensity greater than the separation threshold is extracted as the lane line candidate point.

[0020] Furthermore, in step S5, the removing noise points from the lane line candidate points by using a histogram filtering method includes:

[0021] For the lane line candidate points of the current frame, multiple intervals are evenly divided along the direction perpendicular to the main driving direction, and the number of lane line candidate points in the intervals is counted respectively to generate a point count histogram. The points in the intervals with fewer points are regarded as noise points. A point count threshold is selected to filter out intervals with points lower than the point count threshold.

[0022] Furthermore, in step S6, clustering multiple point cloud points close in position to generate multiple point cloud clusters using the DBSCAN algorithm includes:

[0023] The main driving direction, core radius, search point threshold within the radius, and upper and lower limits of each type of point threshold are set, and the DBSCAN algorithm is used to perform clustering and generate multiple point cloud clusters; the specific process is as follows:

[0024] Define C as a cluster set and N as a candidate set. If the number of points contained in the radius neighborhood of a point exceeds the threshold, the point is considered to be a core point. Select an unprocessed point p. If the point p is the core point, establish a cluster C, and add its neighborhood points to the candidate set N. Select a point q from the candidate set N. If the point q has not been processed and is the core point, add the neighborhood points of the point q to the candidate set N. If the point q has not been classified, add the point q to the cluster C. Repeat the selection of point q until all points in the candidate set N are processed. Repeat the selection of point p until all objects are clustered or classified as noise.

[0025] Furthermore, in step S6, clustering a plurality of point cloud points close in position to generate a plurality of point cloud clusters using the DBSCAN algorithm also includes:

[0026] The distance calculation method in the DBSCAN algorithm is rewritten, and the direction of calculating the distance between points is divided into along the main driving direction and perpendicular to the main driving direction by using the thin and long characteristics of the lane line;

[0027] Low weights and high weights are set for the distances in the directions along the main driving direction and perpendicular to the main driving direction, respectively, and when selecting points that can be clustered into one category, noise points perpendicular to the main driving direction are filtered out.

[0028] Furthermore, in step S7, clustering the plurality of point cloud clusters using a union-find algorithm to cluster the point cloud clusters located on the same lane line into one category includes:

[0029] The center points of the two clusters of point clouds are solved respectively. If the acute angle between the vector formed by the two center points and the vector corresponding to the current driving direction is less than a preset threshold, a connection between the two is established. Then, the union-find algorithm is used to cluster all the point cloud clusters with a connection relationship respectively, and finally multiple point cloud cluster categories located on the same lane line are obtained.

[0030] Furthermore, in step S8, when the lane line points in each category are fitted using the least squares fitting method, if any of the number of points, the difference between the fitting line and the main direction, and the fitting mean square error RMSE does not meet the preset conditions, no fitting is performed.

[0031] Furthermore, in the step S8, after the fitting is completed, a supplementary prediction process is added;

[0032] Using the tracking method, the position change between two frames is obtained through GPS information, and all lane lines detected in the previous frame are projected to the current frame. A candidate area is set near the lane line. If no lane line is generated in the candidate area but there are a certain number of point cloud points, the detection result of the previous frame is used for supplementary prediction.

[0033] Compared with the prior art, the beneficial effects of the present invention include at least:

[0034] 1. The present invention meets the real-time requirements when detecting lane lines, effectively filters out noise points, reduces missed detections and false detections, and ensures accuracy; and the method steps are compact and reasonable, with good verifiability and practicality.

[0035] 2. The present invention uses three-dimensional point clouds to detect lane lines. The lane lines have a higher point cloud intensity because they are coated. By using the property of point cloud intensity, lane line candidate points are extracted from ground points, ensuring their spatial features. This is more accurate than triangulated restoration of the image.

[0036] 3. The present invention proposes a method for solving the adaptive threshold of separating lane line point clouds based on the geometric characteristics of intensity histograms. It fully observes the intensity distribution histograms of multi-frame point clouds, discovers and utilizes the geometric characteristics of the distribution histograms, as well as the characteristics of the relative number and relative intensity of lane line points and ordinary ground points, excludes ordinary ground points, filters out lane line points, and solves the threshold for each frame separately, which is adaptable to various lane line conditions.

[0037] 4. The present invention incorporates multi-frame accumulation, denoising, clustering, tracking and other processing, which effectively improves lane line accuracy, and incorporates a variety of filtering processing to comprehensively improve some difficult-to-process scenes such as dotted lines, arrows, and noise points.

[0038] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0041] Figure 1 A flow chart of a lane line detection method based on laser radar point cloud information provided by the present invention;

[0042] Figure 2 A schematic diagram of a ground point cloud intensity distribution histogram provided by an embodiment of the present invention;

[0043] Figure 3 A schematic diagram of lane line candidate points after noise removal provided by an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of a fitted lane line provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a lane line detection method based on laser radar point cloud information, which at least includes the following steps:

[0047] S1. Obtaining laser radar point cloud information of the vehicle, and extracting point cloud of the area near the ground from the point cloud;

[0048] S2, extracting a ground area from the point cloud of the ground vicinity as a target area;

[0049] S3, performing multi-frame accumulation of ground points in the target area;

[0050] S4, generating an intensity distribution histogram according to the ground points accumulated in multiple frames, obtaining an intensity threshold for extracting the lane line point cloud based on the intensity distribution histogram, and determining the lane line candidate points according to the intensity threshold;

[0051] S5. Using a histogram filtering method to remove noise points from the lane line candidate points;

[0052] S6. Use the DBSCAN algorithm to cluster multiple point cloud points that are close in position to generate multiple point cloud clusters;

[0053] S7, clustering the multiple point cloud clusters using a union-find algorithm, and clustering the point cloud clusters located on the same lane line into one category;

[0054] S8. Use the least squares fitting method to fit each type of lane line point to obtain the lane line equation, and implement lane line detection based on the lane line equation.

[0055] The above steps are described in detail below. The overall method of the embodiment of the present invention is as follows:

[0056] In step 1, the laser radar point cloud information of the vehicle is first obtained, and the point cloud in the rectangular area near the ground is extracted from the point cloud using the rectangular filter in the PCL library.

[0057] In step 2, the ground extraction operation is performed to extract the ground area from the point cloud near the ground as the target area. Since the lane lines are all on the ground, only the ground is taken as a candidate point to reduce the amount of data. The ground extraction can also be completed using the PCL library function.

[0058] In step 3, multiple frames of ground points are accumulated, and the point cloud coordinates are transformed by the change in GPS data between two frames of the ground points. The multi-frame point clouds near the current frame are converted to the current frame coordinate system to enrich the ground points, reduce missed detections, and complement each other between multiple frames.

[0059] In step 4, the adaptive threshold of the point cloud intensity of the extracted lane line candidate points is obtained; each point cloud point has four attributes: x, y, z coordinates and reflection intensity value; for the ground points extracted in the previous step, an intensity distribution histogram is made, and the intensity distribution histogram is as follows Figure 2 As shown, the number of point clouds corresponding to multiple intensity distribution intervals is counted;

[0060] Because of the surface coating, the reflection intensity of the corresponding LiDAR point cloud of the lane line will be significantly higher than that of the ground asphalt points. In addition, due to the special feature that the lane line points are much less than the ground points, the intensity distribution histogram in multiple frames will show similar geometric characteristics: the number of points will be very large at first, and because the reflection intensity of the asphalt points is random, the gradient changes greatly, and it is in the shape of a "wave peak", corresponding to ordinary asphalt points; then as the intensity increases, the number drops sharply, but does not drop to zero. Due to the small number of points and small gradient change, it is in the shape of a "low plain" relative to the first section, corresponding to the lane line candidate points; finally, as the intensity increases, the number of points gradually drops to zero, corresponding to the noise points on the ground from vehicles or special materials;

[0061] What we need to do now is to find the intensity value corresponding to the boundary between the "peak of the twists and turns" and the "low plain". Statistics of multiple frames of LiDAR point clouds show that if the lane line is in good condition, the intensity value can reach more than 15, and even if there is wear, the intensity is at least 8. Using the three characteristics of the small number of lane line points, small gradient, and at least 8 intensity, the intensity is traversed from small to large on the intensity histogram, and the screening conditions are set for the number of points, gradient, and intensity respectively. The first intensity value that meets the three conditions of the number of points less than the point threshold, the gradient less than the gradient threshold, and the intensity greater than the intensity threshold is found as the separation threshold to be obtained, and the threshold is repeatedly obtained for each frame to achieve an adaptive effect for each frame. If the lane line wear changes between multiple frames, it can also be accurately extracted. Point clouds with an intensity greater than the separation threshold will be extracted as lane line candidate points.

[0062] In step 5, the lane line candidate points are denoised. Due to the special geometric shape of the lane lines, which have obvious thin and long features, and the fact that most lane lines are parallel to each other and the driving direction, the histogram filtering method is used to remove noise points.

[0063] First, the main driving direction is obtained. Using the correlation between the previous and next frames, based on the detection results of the previous frame, a lane line with the smallest mean square error and sufficient number of points is found among all the detected lane lines. Its slope is used as the main driving direction of the current frame. The lane line candidate points of the current frame are evenly divided into multiple intervals along the direction perpendicular to the main driving direction. The number of lane line candidate points in each interval is counted and a point count histogram is made. The interval that can generate lane lines must have a large number of points, and the histogram will be "multi-peaked". Each lane line corresponds to a local maximum point count in the corresponding area.

[0064] Then, we take a point count threshold, retain the "multi-peaks", and filter out intervals with points below the threshold. That is, the points in the intervals with fewer points are considered as noise points. This operation can effectively remove noise points from the lane line candidate points. The lane line candidate points after denoising are as follows: Figure 3 Shown as a dark point cloud.

[0065] In step 6, the DBSCAN algorithm is used to cluster multiple point cloud points that are close in position, and finally a grouping effect is presented. Points located on the same continuous lane line will be clustered into one category.

[0066] The DBSCAN algorithm defines a cluster as the largest set of density-connected points. It can divide areas with sufficiently high density into clusters and can find clusters of arbitrary shapes in noisy spatial databases.

[0067] The algorithm process is C as the cluster set and N as the candidate set. Select an unprocessed point p. If the number of points contained in the radius neighborhood of a point exceeds the threshold, the point is considered to be a core point. If p is a core point, establish a cluster set C, and add its neighborhood points to the candidate set N. Select point q from N. If q has not been processed and is a core point, add q's neighborhood points to N. If q has not been classified, add q to C. Repeat the selection of q until all points in N have been processed. Repeat the selection of p until all objects are clustered or classified as noise. The multiple clusters generated at this time are the clustering results.

[0068] In this step, the distance calculation method in the DBSCAN radius search link is also rewritten. By utilizing the thin and long characteristics of lane lines, the directions of solving the distance between points are divided into along the main driving direction and perpendicular to the main driving direction. The distances in these two directions are set with low weights and high weights respectively, so that when selecting points that can be clustered into one category, the noise points perpendicular to the main driving direction can be effectively filtered out and the lane line points perpendicular to the main driving direction can be retained.

[0069] The main driving direction, core radius, radius search point threshold, and upper and lower thresholds of each type of points are set, and the DBSCAN algorithm is used for clustering to generate multiple point cloud clusters.

[0070] In step 7, the union-find algorithm is used to cluster the major categories. For discontinuous lane lines, such as dotted lines, multiple groups of point clouds on the same lane line should be clustered into one category to avoid repeated fitting and to more accurately describe the position and slope characteristics of the lane line. It can also solve the problem that the point clouds are dense at the beginning and sparse at the end and cannot be clustered into one category.

[0071] Union-find is a tree-type data structure used to handle the merging and querying of some disjoint sets; it is divided into three steps: initialization, search, and merging.

[0072] First, initialize the set of each point to itself, then find the set where the element is located, that is, the root node, and finally merge the sets of the two elements into one set, that is, establish a connectivity relationship.

[0073] Generally speaking, before merging, you should first determine whether the two elements belong to the same set, which can be achieved using the "find" operation above.

[0074] The relationship between each element in the set is determined by judging the proximity to the main direction. For two clusters of point clouds, the following method is used to determine whether there is a relationship: solve the center points of the two clusters of point clouds respectively. If the acute angle between the vector formed by the two center points and the vector corresponding to the current driving direction is small, it means that the two clusters of point clouds may be located on the same lane line and can be clustered into one category. Using the relationship between the two clusters, the derivative groups of each cluster of lane line point clouds are found to achieve the effect of clustering small lane lines into large categories.

[0075] In step 8, the lane line equation is fitted using least squares; a set of point cloud coordinates are input and the corresponding parameters of a straight line that minimizes the sum of the distances from the points to the straight line are found.

[0076] During fitting, continue with a round of filtering. If there are too many or too few points of a certain type, it means that the clustering is unreliable and no fitting is done. If the difference between the fitting line and the main direction is large or the fitting mean square error (RMSE) is large, no fitting is done.

[0077] After fitting is completed, a supplementary prediction process is added. Using the tracking method, the position change between the two frames is obtained through GPS information. All lane lines detected in the previous frame are projected to the current frame. A candidate area is set near the lane line. If there is no lane line generated in the candidate area but there are a certain number of point cloud points, it means that there is a missed detection. The detection result of the previous frame can be used for supplementary prediction. Finally, a richer lane line equation is obtained. After fitting, the lane line equation is visualized as follows Figure 4 As shown, the dark points in the point cloud are the extracted lane line candidate points, and the light square points are the lane line points after the fitting equation is visualized.

[0078] After all equations are fitted, filter again.

[0079] Find the ones with incorrect intervals and filter them out. The interval is generated by using the previous frame data to find the first lane line on the left and right of the vehicle and calculate their distance. As the interval prior, the arrow is easily misdetected as a dotted lane line during point cloud lane line detection. The lane line with the smallest mean square error and the largest number of points is used as the benchmark lane line. Then, the interval can be used to filter out the arrow to eliminate the interference of the arrow lane line.

[0080] Extract the curb candidate points and filter out the lane lines outside the curb. The curb features are used when extracting the curb. The ground point cloud corresponding to the road surface cut off by the curb will also be cut off, which is very obvious. Therefore, the boundary where the ground point cloud suddenly disappears can be used as a curb reference.

[0081] The present invention provides a lane line detection method based on laser radar point cloud information, which selects lane line candidate points through the reflection intensity attribute of the point cloud; proposes a method for solving the adaptive threshold of extracting lane line point cloud based on the point cloud intensity histogram, and uses the geometric characteristics of the distribution, the characteristics of the relative number of lane line points and ordinary ground points to exclude ordinary ground points, and the practical effect is better than other schemes. And add multi-frame accumulation, denoising, clustering, tracking, curve fitting and other processing, and finally obtain the lane line equation. The present invention utilizes the characteristics that point clouds have rich three-dimensional information and high position accuracy in lane line detection, which is more accurate than image triangulation recovery, and can enrich ground points by using the intensity distribution curve characteristics to solve the threshold, multi-frame accumulation and union-find algorithm clustering and processing and other links, and can solve the problem of laser radar point cloud being dense in front and sparse in the back, and dotted lane lines being easy to miss detection; make full use of the various set characteristics of lane lines being slender, parallel to each other, and parallel to the main direction for filtering and screening; improve the detection accuracy of lane lines, and the detection is highly real-time.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A lane line detection method based on laser radar point cloud information, characterized in that: include: S1. Obtaining laser radar point cloud information of the vehicle, and extracting point cloud of the area near the ground from the point cloud; S2, extracting a ground area from the point cloud of the ground vicinity as a target area; S3, performing multi-frame accumulation of ground points in the target area; S4, generating an intensity distribution histogram according to the ground points accumulated in multiple frames, obtaining an intensity threshold for extracting the lane line point cloud based on the intensity distribution histogram, and determining the lane line candidate points according to the intensity threshold; S5. Using a histogram filtering method to remove noise points from the lane line candidate points; S6. Use the DBSCAN algorithm to cluster multiple point cloud points that are close in position to generate multiple point cloud clusters; S7, clustering the multiple point cloud clusters using a union-find algorithm, and clustering the point cloud clusters located on the same lane line into one category; S8. Fit each type of lane line point using a least squares fitting method to obtain a lane line equation, and implement lane line detection based on the lane line equation; In step S6, clustering multiple point cloud points close in position to generate multiple point cloud clusters using the DBSCAN algorithm includes: The distance calculation method in the DBSCAN algorithm is rewritten, and the direction of calculating the distance between points is divided into along the main driving direction and perpendicular to the main driving direction by using the thin and long characteristics of the lane line; Low weights and high weights are set for the distances in the directions along the main driving direction and perpendicular to the main driving direction, respectively, and when selecting points that can be clustered into one category, noise points perpendicular to the main driving direction are filtered out.

2. The lane line detection method based on laser radar point cloud information as claimed in claim 1, characterized in that: The step S1 specifically includes: obtaining the laser radar point cloud information of the vehicle, and using the cuboid filter in the PCL library to extract the point cloud in the cuboid area near the ground from the point cloud information.

3. The lane line detection method based on laser radar point cloud information as claimed in claim 2, characterized in that: In the step S3, multi-frame accumulation of ground points in the target area includes: By using the change in GPS data between two frames of the ground point, the point cloud coordinates are transformed, and multiple frame point clouds near the current frame are converted to the current frame coordinate system, so that the multiple frames complement each other.

4. The lane line detection method based on laser radar point cloud information as claimed in claim 3, characterized in that: In the step S4, obtaining an intensity threshold for extracting a lane line point cloud based on the intensity distribution histogram, and determining lane line candidate points according to the intensity threshold, includes: Traversing the intensity of the point cloud from small to large on the intensity distribution histogram, setting filtering conditions for the number of points, gradient, and intensity respectively, and finding the first intensity value that meets the three conditions that the number of points is less than the point number threshold, the gradient is less than the gradient threshold, and the intensity is greater than the intensity threshold, as the separation threshold to be obtained; For each frame, the separation threshold is repeatedly obtained, and the point cloud with intensity greater than the separation threshold is extracted as the lane line candidate point.

5. The lane line detection method based on laser radar point cloud information as claimed in claim 4, characterized in that: In step S5, the use of a histogram filtering method to remove noise points from the lane line candidate points includes: For the lane line candidate points of the current frame, multiple intervals are evenly divided along the direction perpendicular to the main driving direction, and the number of lane line candidate points in the intervals is counted respectively to generate a point count histogram. The points in the intervals with fewer points are regarded as noise points. A point count threshold is selected to filter out intervals with points lower than the point count threshold.

6. The lane line detection method based on laser radar point cloud information as claimed in claim 5, characterized in that: In step S6, clustering multiple point cloud points close in position to generate multiple point cloud clusters using the DBSCAN algorithm includes: The main driving direction, core radius, search point threshold within the radius, and upper and lower limits of each type of point threshold are set, and the DBSCAN algorithm is used to perform clustering and generate multiple point cloud clusters; the specific process is as follows: Define C as a cluster set and N as a candidate set. If the number of points contained in the radius neighborhood of a point exceeds the threshold, the point is considered to be a core point. Select an unprocessed point p. If the point p is the core point, establish a cluster C, and add its neighborhood points to the candidate set N. Select a point q from the candidate set N. If the point q has not been processed and is the core point, add the neighborhood points of the point q to the candidate set N. If the point q has not been classified, add the point q to the cluster C. Repeat the selection of point q until all points in the candidate set N are processed. Repeat the selection of point p until all objects are clustered or classified as noise.

7. The lane line detection method based on laser radar point cloud information as claimed in claim 1, characterized in that: In step S7, clustering the plurality of point cloud clusters using a union-find algorithm to cluster the point cloud clusters located on the same lane line into one category includes: The center points of the two clusters of point clouds are solved respectively. If the acute angle between the vector formed by the two center points and the vector corresponding to the current driving direction is less than a preset threshold, a connection between the two is established. Then, the union-find algorithm is used to cluster all the point cloud clusters with a connection relationship respectively, and finally multiple point cloud cluster categories located on the same lane line are obtained.

8. The lane line detection method based on laser radar point cloud information as claimed in claim 7, characterized in that: In step S8, when fitting each type of lane line points using the least squares fitting method, if any of the number of points, the difference between the fitting line and the main direction, and the fitting mean square error RMSE does not meet the preset conditions, no fitting is performed.

9. The lane line detection method based on laser radar point cloud information as claimed in claim 8, characterized in that: In the step S8, after the fitting is completed, a supplementary prediction process is added; Using the tracking method, the position change between two frames is obtained through GPS information, and all lane lines detected in the previous frame are projected to the current frame. A candidate area is set near the lane line. If no lane line is generated in the candidate area but there are a certain number of point cloud points, the detection result of the previous frame is used for supplementary prediction.

Images