WO2023124231A1 - 基于激光雷达的道路边界检测方法和装置 - Google Patents

基于激光雷达的道路边界检测方法和装置 Download PDF

Info

Publication number
WO2023124231A1
WO2023124231A1 PCT/CN2022/118213 CN2022118213W WO2023124231A1 WO 2023124231 A1 WO2023124231 A1 WO 2023124231A1 CN 2022118213 W CN2022118213 W CN 2022118213W WO 2023124231 A1 WO2023124231 A1 WO 2023124231A1
Authority
WO
WIPO (PCT)
Prior art keywords
candidate
grids
grid
point cloud
road
Prior art date
Application number
PCT/CN2022/118213
Other languages
English (en)
French (fr)
Inventor
庞伟凇
王宇
周琳
林崇浩
Original Assignee
中国第一汽车股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国第一汽车股份有限公司 filed Critical 中国第一汽车股份有限公司
Publication of WO2023124231A1 publication Critical patent/WO2023124231A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads

Definitions

  • the embodiments of the present application relate to the field of vehicles, and in particular, relate to a method and device for detecting road boundaries based on lidar.
  • the detection of road boundaries is the core functional component that provides a safe driving area for humans in assisted driving, and is also an important prerequisite for providing effective detection range and drivable area for fully automatic driverless driving.
  • Set functional components are the core functional component that provides a safe driving area for humans in assisted driving, and is also an important prerequisite for providing effective detection range and drivable area for fully automatic driverless driving.
  • road edge detection algorithms are mostly based on mechanical lidar point clouds.
  • solid-state lidar has emerged as the times require, and is in mass production.
  • Self-driving cars are gaining ground. Due to the different scanning methods, the point clouds formed by solid-state lidar and mechanical lidar are also very different, so there is a problem that road detection algorithms cannot be well applied to solid-state lidar scenarios.
  • Embodiments of the present application provide a lidar-based road boundary detection method and device to at least solve the technical problem that current road detection algorithms cannot be applied to solid-state lidar scenarios.
  • a road boundary detection method based on laser radar including: scanning the road image of the structured road to obtain the laser radar point cloud data of the structured road; Rasterization processing, generating multiple grids, and obtaining the grid feature information of the point cloud in each grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids , to obtain multiple target candidate grids; perform area marking on multiple target candidate grids to obtain at least one roadside candidate area of the structured road; based on the area candidate points in each roadside candidate area, generate the structured road road boundary.
  • At least one screening process is performed on multiple grids, including: using at least one screening condition to filter the grid feature information of the point cloud in each grid Perform screening processing, wherein the screening conditions include at least one of the following: a first screening condition, wherein the screening factors included in the first screening condition include: the height of the highest point in the grid, the lowest point in the grid and the height difference of the point cloud in the grid; the second filter condition, wherein the filter factors included in the second filter condition include: whether the number of candidate grids around the current candidate grid is within the range of the screening threshold.
  • using the first filtering condition to filter the grid feature information of the point cloud in each grid includes: using one or more filtering factors in the first filtering condition to filter the point cloud in each grid
  • the raster feature information is screened to obtain the first group of candidate rasters, where the point clouds in the first group of candidate rasters include the following features: no false edges, low obstacles, and point clouds with height difference objects data.
  • the method further includes: using a second screening condition to perform secondary screening on the candidate grids in the first group of candidate grids to obtain the second group of candidate grids, wherein, The second group of candidate grids is the candidate grids in the first group of candidate grids that have undergone secondary grid marking.
  • the second screening condition is used to perform secondary screening on the candidate grids in the first group of candidate grids to obtain the second group of candidate grids, including: detecting the position of each candidate grid in the first group of candidate grids The number of candidate grids in their respective neighborhoods, where the candidate grids are the grids in the first group of candidate grids; if the number of candidate grids in the neighborhood of any candidate grid in the first group of candidate grids is within the screening threshold range , then the candidate rasters within the screening threshold range will be re-marked as rasters.
  • performing region marking on multiple target candidate grids to obtain at least one roadside candidate region of the structured road including: using a search algorithm to search for the second group of candidate grids Merge the candidate grids, and count the number of candidate grids in each candidate area after merging; mark the second group of candidate grids that have been merged, and obtain at least one roadside candidate area, where, The roadside candidate area includes multiple target candidate grids.
  • the candidate area is marked as an area.
  • generating the road boundary of the structured road includes: obtaining multiple grids in any roadside candidate area; selecting candidate points from multiple grids , to obtain the candidate points in each roadside candidate area; perform curve fitting on all the candidate points in each roadside candidate area, wherein the fitted curve is the road boundary of the structured road.
  • the method further includes: preprocessing the lidar point cloud data, wherein the preprocessing includes at least one of the following: point cloud undefined value filtering, noise point filtering as well as point cloud height filtering.
  • a lidar-based road boundary detection device including: an acquisition component, configured to scan the road image of the structured road, and acquire the lidar point cloud data of the structured road ;
  • the processing component is set to rasterize the lidar point cloud data, generate multiple grids, and obtain the grid feature information of the point cloud in each grid;
  • the filtering component is set to be based on each grid
  • the raster feature information of the point cloud performs at least one screening process on multiple rasters to obtain multiple target candidate rasters;
  • the marking component is set to perform area marking on multiple target candidate rasters to obtain at least one of the structured roads A roadside candidate area;
  • a generation component configured to generate an overall error of the target model based on the detection error of the original model and the detection error of the target model.
  • a non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the lidar-based road boundary detection method of the embodiment of the present application.
  • a processor is also provided.
  • the processor is configured to run a program, wherein the detection of the road boundary based on lidar according to the embodiment of the present application is performed when the program is running.
  • the road image of the structured road is scanned to obtain the laser radar point cloud data of the structured road; the laser radar point cloud data is rasterized to generate multiple grids, and each grid is obtained The grid feature information of the point cloud in each grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple target candidate grids; perform multiple target candidate grids area marking, at least one roadside candidate area of the structured road is obtained; based on area candidate points in each roadside candidate area, a road edge of the structured road is generated.
  • Fig. 1 is a flow chart of a method for detecting road boundaries based on lidar according to an embodiment of the present application
  • FIG. 2 is a flowchart of another laser radar-based road boundary detection method according to an embodiment of the present application.
  • FIG. 3 is a schematic diagram of a roadside distribution according to an embodiment of the present application.
  • FIG. 4 is a schematic diagram of a lidar-based road boundary detection method device according to an embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of a non-volatile storage medium according to an embodiment of the present application.
  • Fig. 6 is a schematic structural diagram of a processor according to an embodiment of the present application.
  • an embodiment of a road boundary detection method based on laser radar is provided. It should be noted that the steps shown in the flow chart of the accompanying drawings can be implemented in a computer system such as a set of computer-executable instructions and, although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that shown or described herein.
  • Fig. 1 is a flow chart of a road boundary detection method based on laser radar according to an embodiment of the present application, as shown in Fig. 1 , a flow chart of a road boundary detection method based on laser radar, the method includes the following steps:
  • Step S102 scanning the road image of the structured road to obtain laser radar point cloud data of the structured road.
  • the road image of the structured road is scanned to obtain the laser radar point cloud data of the structured road.
  • the structured road can be a relatively regular structured roadside in an urban road, and it is also possible to detect roadside bushes or water horses and temporary baffles that appear in temporary road construction.
  • Unstructured roads such as rural roads that do not have obvious and clear road boundaries are not detection targets.
  • the laser radar point cloud data can be used by the laser radar sensor to measure the propagation distance between the sensor emitter and the target object, analyze the size of the reflected energy on the surface of the object, the amplitude, frequency and phase of the reflected spectrum, and so on. It presents the 3D point cloud data of the data collected during the driving process of the self-driving vehicle, where the lidar sensor can be a binocular camera, a 3D scanner, etc. For example, scan the graphics captured by the lidar sensor, and obtain the lidar point cloud data based on the intrinsic parameters of the camera.
  • Step S104 performing rasterization processing on the lidar point cloud data to generate multiple rasters, and acquiring raster feature information of the point cloud in each raster.
  • the laser radar point cloud data is rasterized to generate multiple grids, and the statistics are collected during the rasterization process.
  • the feature information of the point cloud in the raster is collected.
  • the rasterization process may be converting the graphics represented by the vector graphic format into a grid graphic, and may be dividing the point cloud into multiple grids in the two directions of the horizontal coordinate axis.
  • the characteristic information of each grid is counted, and the characteristic information can be the point cloud height difference in the grid, the maximum height of the point cloud in the grid, the minimum height of the point cloud in the grid, etc.
  • the point cloud height difference in the grid can be represented by Zdiff
  • the maximum height of the point cloud in the grid can be represented by Zgrid_max
  • the minimum height of the point cloud in the grid can be represented by Zgrid_min.
  • the point cloud is evenly divided into m*n grids in the xy direction, and the three feature information of each grid are counted during the division process, including the height difference of the point cloud in the grid, the point cloud in the grid Maximum height, the minimum height of the point cloud within the raster.
  • Step S106 based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple target candidate grids.
  • step S106 of the present application based on the obtained grid feature information of the point cloud in each grid, and according to the set filtering conditions, the grid is screened once to obtain the target candidate grid .
  • the filtering process can be to filter out grids that do not meet the conditions according to the set conditions, and the screening process is performed at least once, wherein the filtering condition can be at least one condition set according to the feature information , for example, set the maximum height of the point cloud in the grid, and when the statistical feature information is greater than the maximum height of the point cloud in the set grid, filter out this grid to obtain the target candidate grid, or/and, Set the minimum height of the point cloud in the grid. When the statistical feature information is less than the minimum height of the point cloud in the set grid, filter out this grid to obtain the target candidate grid, or/and, set the grid The maximum height difference of the point cloud in the grid.
  • the filtering condition can be at least one condition set according to the feature information , for example, set the maximum height of the point cloud in the grid, and when the statistical feature information is greater than the maximum height of the point cloud in the set grid, filter out this grid to obtain the target candidate grid, or/and, Set the minimum height of the point cloud in the grid.
  • this grid is filtered out to obtain the target candidate grid, or/and, the set grid The minimum height difference of the internal point cloud.
  • the grid is filtered out to obtain the target candidate grid.
  • the target candidate grid may be a grid obtained after rasterizing the lidar point cloud data of the obtained structured road and filtering out unqualified grids according to set conditions.
  • Step S108 region marking is performed on a plurality of target candidate grids to obtain at least one roadside candidate region of the structured road.
  • region marking is performed on multiple target candidate grids, that is, all centroids in the target candidate grids are selected as candidate points and region marked, so as to obtain at least one path of the structured road. Along the candidate area.
  • Step S110 based on the area candidate points in each roadside candidate area, generate a road boundary of the structured road.
  • step S110 of the present application all candidate points in the candidate area are used for curve fitting, combined with the actual situation of the road boundary, the road edge curve is finally fitted, and the road boundary of the structured road is generated.
  • step S102 to step S110 of the present application scans the road image of the structured road to obtain the laser radar point cloud data of the structured road; rasterizes the laser radar point cloud data to generate multiple grids, and obtains each The grid feature information of the point cloud in the grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple target candidate grids; for multiple target candidate grids
  • the grid is marked to obtain at least one roadside candidate area of the structured road; based on the area candidate points in each roadside candidate area, the road edge of the structured road is generated.
  • step S106 is to perform at least one screening process on multiple grids based on the grid feature information of the point cloud in each grid.
  • the method also includes: using at least one screening condition to The grid feature information of the point cloud in each grid is screened, wherein the filter conditions include at least one of the following: a first filter condition, wherein the filter factors included in the first filter condition include: the height of the highest point in the grid , the height of the lowest point in the grid and the height difference of the point cloud in the grid; the second filter condition, wherein the filter factors included in the second filter condition include: whether the number of candidate grids around the current candidate grid is within the range of the screening threshold Inside.
  • the acquired lidar point cloud data is rasterized to generate a plurality of raster information, and at the same time, the raster feature information of the point cloud in each raster is counted, when the feature information satisfies the first filtering condition , mark the grid as a candidate grid, and other grids that do not meet the first filtering condition will not be marked, and will not be processed later. Then, it is judged whether the marked candidate grid meets the second filtering condition. For Grids that meet the second screening condition are marked to obtain target candidate grids.
  • the filtering factors contained in the first filtering condition may include: the height of the highest point in the grid, the height of the lowest point in the grid, the maximum height difference of the point cloud in the grid, and the minimum height of the point cloud in the grid.
  • the height difference wherein, the height of the highest point in the grid, the height of the lowest point in the grid and the height difference of the point cloud in the grid can be values obtained through multiple tests, which are set by the user and input into the system.
  • the screening factors included in the second screening condition may include whether the number of candidate grids around the current candidate grid is within a screening threshold range.
  • the second filtering condition is used to filter out the number of candidate grids around the current candidate grid.
  • grid where the maximum and minimum values of the number of test candidate grids are set, so as to filter out too many grid candidate grids.
  • the target candidate grid is obtained.
  • using the first filtering condition to filter the grid feature information of the point cloud in each grid includes: using one or more filtering factors in the first filtering condition, for each The grid feature information of the point cloud in the first grid is screened to obtain the first group of candidate grids, wherein the point cloud in the first group of candidate grids includes the following features: no false edges, low obstacles, and Point cloud data of height difference objects.
  • one or more screening factors in the first screening condition are used to filter the grid feature information of the point cloud in each grid to obtain the first group of candidate grids, wherein the first screening The conditions can be: the height difference in the grid feature information cannot be greater than the maximum height difference of the point cloud in the set grid; the height difference in the grid feature information cannot be less than the minimum height of the point cloud in the set grid Poor; the height of the highest point in the raster feature information cannot be less than the height of the lowest point in the raster; the height of the lowest point in the raster feature information cannot be greater than the height of the highest point in the raster.
  • the first candidate data is obtained by screening.
  • the maximum height difference of the point cloud in the grid can be represented by max_threshold, and the height difference in the grid feature information cannot be greater than the set maximum height difference of the point cloud in the grid, that is, Zdiff ⁇ max_threshold, thus filtering False edges such as falling off tall buildings.
  • the minimum height difference of the point cloud in the grid can be represented by min_threshold, and the height difference in the grid feature information cannot be less than the minimum height difference of the point cloud in the set grid, that is, Zdiff>min_threshold, thereby filtering out the ground and the ground Low obstacles such as small protrusions.
  • the height of the lowest point in the grid can be represented by min_height, and the height of the highest point in the grid feature information cannot be less than the height of the lowest point in the grid, that is, Zgrid_max>min_height, so as to filter out some special low obstacles.
  • the height of the highest point in the grid can be represented by max_height.
  • the height of the lowest point in the grid feature information cannot be greater than the height of the highest point in the grid, so as to filter out some objects that are far from the ground or have height differences in the air.
  • the maximum height difference of the point cloud in the grid For example, set the maximum height difference of the point cloud in the grid to 3.0 meters during the test.
  • the minimum height difference of the point cloud in the grid used for the test is set to 0.10 meters, when the height in the grid feature information
  • the difference is greater than the minimum height difference of the point cloud in the set grid, it will be filtered out and the qualified grid will be marked; the height of the lowest point in the grid is set to 0.05 meters during the test, and the highest point in the grid feature information is 0.05 meters.
  • the manner of marking the grid may be marking with colors, graphics, etc., which is not limited here.
  • the method further includes: using the second filtering condition to perform secondary screening on the candidate grids in the first group of candidate grids, and obtain the second A group of candidate grids, wherein the second group of candidate grids is the candidate grids in the first group of candidate grids that have undergone secondary grid marking.
  • the first group of candidate grids is obtained and marked, the first group of marked candidate data is screened and marked by the second filtering condition, and the grids with two candidate marks are selected as the second group of candidate grids grid, where the mark can be color, graphics, etc., which is not limited here.
  • the second screening condition is used to perform secondary screening on the candidate grids in the first group of candidate grids to obtain the second group of candidate grids, including: detecting The number of candidate grids in their respective neighborhoods for each candidate grid, where the candidate grid is the grid in the first group of candidate grids; if any candidate grid in the first group of candidate grids is in the neighborhood of candidate grids is within the range of the screening threshold, the candidate rasters within the range of the screening threshold are marked for secondary rasterization.
  • the candidate grids in the first group of candidate grids are screened twice by using the second screening condition to obtain the second group of candidate grids.
  • the screening factors included in the second screening condition may include whether the number of candidate grids around the current candidate grid is within a screening threshold range.
  • the first group of candidate grids obtained through the first screening condition count the number of candidate grids in the 8 neighborhoods of the first group of candidate grids, where the candidate areas can be up, down, left, upper left, lower left, upper right, and lower right.
  • the currently selected grid is a candidate grid
  • the number of candidate grids It can be represented by neighbor.
  • the second filter condition is used to filter out the grids with too many candidate grids around the current candidate grid, that is, the maximum number of candidate grids is greater than or equal to the minimum number of candidate grids and less than Equal to the maximum number of candidate grids, where the maximum number of candidate grids can be represented by Max_neighbor, and the minimum number of candidate grids can be represented by Min_neighbor.
  • step S108 after obtaining the second group of candidate grids, performs area marking on multiple target candidate grids to obtain at least one roadside candidate area of the structured road, including: using search The algorithm merges the candidate grids in the second group of candidate grids, and counts the number of candidate grids in each candidate area after merging; marks the area of the second group of candidate grids that have been merged, and obtains at least A roadside candidate area, wherein the roadside candidate area includes a plurality of target candidate grids.
  • a search algorithm is used to merge candidate grids in the second group of candidate grids, and count the number of candidate grids in each candidate area after merging; for the second group of candidate grids that have been merged Grids are used to mark regions and obtain multiple target candidate regions.
  • the search algorithm can randomly select a first group of candidate grids as seed points, create a new set of candidate grids to store the grids, and determine whether there are candidate grids in the field of seed point 8, and if so, put them into the set , until all the grids connected with the seed point are put into the set, reselect a candidate grid that has not been visited as the seed point, create a new set of candidate grids again, and repeat the above operations to obtain multiple candidate areas Multiple candidate raster sets for . Count the number of each candidate raster set, where the number of candidate raster sets can be represented by Ncandidate.
  • search algorithm is not specifically limited, and the number of candidate grid sets can be counted by other search algorithms.
  • the candidate area is marked as an area.
  • the target threshold can be the value set by the system, represented by Min_candidate, because under normal circumstances, road boundaries appear continuously, so when the number of grids in the candidate area is too small, it is determined to be a false detection. Subsequent processing as road boundaries.
  • the target threshold is set to 15 during testing, the number of candidate raster sets must be greater than the target threshold, that is, Ncandidate>Min_candidate.
  • the candidate raster set will not be selected. As a road boundary, no subsequent processing is performed.
  • step S110 based on the area candidate points in each roadside candidate area, generates the road boundary of the structured road, including: acquiring multiple grids in any roadside candidate area; Select candidate points from multiple grids to obtain candidate points in each roadside candidate area; perform curve fitting on all candidate points in each roadside candidate area, wherein the fitted curve is a structured road road boundaries.
  • multiple grids in any one roadside candidate area are obtained; candidate points are selected from multiple grids, and candidate points in each roadside candidate area are obtained; for each roadside candidate area Curve fitting is performed on all the candidate points to obtain the road boundary of the structured road.
  • selecting candidate points from multiple grids can be selected as the center of gravity of all points in the grid as candidate points, thereby reducing the amount of calculation, combined with the actual situation of the road boundary, selecting an appropriate fitting method for the candidate points used Curve fitting is performed on all candidate points in the area to obtain the road boundary of the structured road.
  • the fitting method is a fitting method that can complete curve fitting, such as least square method, random sampling consensus algorithm, etc., which is not specifically limited here.
  • selecting candidate points from multiple grids can be to select the center of gravity of all points in the grid as candidate points, and use a cubic curve equation for fitting.
  • the fitting method uses a random sampling consensus algorithm, and the final fitted curve is the road edge curve.
  • the method further includes: preprocessing the lidar point cloud data, wherein the preprocessing includes at least one of the following: point cloud Undefined value filtering, noise point filtering and point cloud height filtering.
  • the preprocessing of the lidar point cloud data may mainly include: point cloud undefined value filtering, noise point filtering and point cloud height filtering.
  • point cloud undefined value filtering is used to filter out points with abnormal point cloud coordinates in the original point cloud data, which can be done by traversing the point cloud and judging the point cloud coordinates; noise point filtering can be done by rasterizing the point cloud , divide the point cloud from three dimensions into two-dimensional grids, count the number of point clouds in the grid, and set a threshold according to the actual situation.
  • point cloud height filtering can filter out point clouds whose absolute height of point cloud is too high or too low.
  • the road image of the structured road is scanned to obtain the laser radar point cloud data of the structured road; the laser radar point cloud data is rasterized to generate multiple grids, and the The grid feature information of the point cloud in each grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple target candidate grids; for multiple targets
  • the candidate grid performs area marking to obtain at least one roadside candidate area of the structured road; based on the area candidate points in each roadside candidate area, the road edge of the structured road is generated.
  • autonomous driving provides more possibilities for liberating human hands.
  • autonomous driving whether it is L2, L3 and other assisted driving or L4, L5 and other more advanced automatic driving, it is inseparable from the detection of road boundaries. It is the core functional component in assisted driving that provides a safe driving area for humans. , It is also an important front-end functional component that provides effective detection range and drivable area for fully automatic driverless driving.
  • the premise of this method is to obtain the point cloud of the same scanning line, that is, the beam information of the radar is required, but for a frame of point cloud, the coordinates and intensity of the points are the basic information, while the laser beam is additional.
  • solid-state lidar is not a scanning line fan scan, and most of them do not have the so-called beam information, which makes the detection method based on the wire beam information unsuitable;
  • the detection algorithm relies on the structured road model, and the detection object is relatively single.
  • most detection algorithms were based on very idealized structured road models. Most of the detection objects were relatively formal structured roadsides in urban roads. For roadside bushes or water horses and temporary baffles in temporary road construction, etc. Neither can be detected.
  • a road boundary detection method based on three-dimensional laser radar is designed, and the obstacle and non-obstacle are discriminated by using the height difference of the grid point cloud
  • the distance grayscale image is analyzed, and the area outline is obtained by using the area growing method, but this algorithm still has the problem of being unable to detect the roadside bushes or the water horses and temporary baffles that appear in the temporary construction of the road. .
  • a road edge detection method and device which uses the normal and normal curvature of the point cloud to extract road edge candidate points, and then uses region growing and concave packet algorithms to remove noise points,
  • the algorithm still has the problem of being unable to detect roadside bushes or water horses and temporary baffles that appear in temporary road construction.
  • a real-time road boundary detection method based on the local concave-convex features of the point cloud is proposed, which converts the point cloud into a depth-expanded image, and then uses the concave-convex features of the image for boundary detection, thereby realizing road detection,
  • this method has the problem that it is not suitable for solid-state lidar point clouds.
  • a method and device for real-time extraction and measurement of road boundaries are proposed.
  • the method uses points with sudden changes in point cloud angles as boundary candidate points, and then separates the left and right sides Construct a mathematical description model to improve the speed of algorithm operation in the road detection process, but this method has the problem that it is not suitable for solid-state lidar point clouds.
  • a road boundary detection method based on lidar is proposed.
  • Carry out road boundary detection not only can detect regular road edges in structured roads, but also can detect road boundaries formed by roadside bushes and temporary road boundaries formed by water horses and temporary baffles placed by road construction, and
  • the algorithm of the embodiment of the present application has high real-time performance, and a detection model suitable for both mechanical lidar point cloud and solid-state lidar point cloud is designed.
  • Figure 2 is a flow chart of a road boundary detection method based on laser radar according to an embodiment of the present application, and the method may include the following steps:
  • Step S201 preprocessing the point cloud.
  • the point cloud is preprocessed.
  • the point cloud preprocessing in the detection process it mainly includes three parts: point cloud uncertainty value filtering, noise point filtering and point cloud height filtering.
  • uncertain value filtering refers to filtering out points with abnormal point cloud coordinates in the original point cloud, which can be completed by traversing the point cloud and judging the point cloud coordinates; noise point filtering is completed by rasterizing the point cloud, that is, point cloud from xyz
  • the three dimensions are divided into m*n*l grids, and the number of point clouds in the grid is counted.
  • N can be set according to the actual situation of the user, this application
  • N can be set according to the actual situation of the user, this application
  • N can be set according to the actual situation of the user, this application
  • N can be set according to the actual situation of the user, this application
  • N can be set according to the actual situation of the user, this application
  • N can be set according to the actual situation of the user, this application
  • Cloud height filtering only refers to filtering out point clouds whose absolute height is too high or too low, because the height will not be too high or too low for the edge of the road, so if the point cloud height direction coordinate z satisfies z ⁇ Zmin or z >Zmax
  • the setting of Zmin and Zmax needs to be determined in combination with the actual situation of the user, that is, the installation height of the radar and the origin location of the point cloud coordinate system.
  • Step S202 format the point cloud grid, and collect grid feature information.
  • the processed points are rasterized, and at the same time, the feature information of the point cloud in the raster is counted during the rasterization process.
  • the point cloud described in the detection process is rasterized and the point cloud information in the grid is counted. Specifically, the point cloud is divided into m*n grids in the xy direction, and the three grids of each grid are counted during the division process.
  • Feature information including the point cloud height difference Zdiff in the grid, the maximum height Zgrid_max of the point cloud in the grid, and the minimum height Zgrid_min of the point cloud in the grid.
  • Step S203 judging whether the data satisfies condition 1.
  • condition 1 in the detection process specifically refers to: 1) Zdiff ⁇ max_threshold, that is, the point cloud height difference in the grid should not be too large, so as to filter out false edges such as tall buildings, the embodiment of the present application
  • max_threshold 3.0m
  • Zdiff>min_threshold that is, the height difference of the point cloud in the grid should not be too small, so as to filter out low obstacles such as the ground and small protrusions on the ground.
  • Step S204 marking the grid as a candidate grid.
  • Step S205 counting the number of candidate grids in the field of candidate grid 8.
  • For candidate grids count the number of candidate grids in each grid's 8-neighborhood.
  • the counting of the number of candidate grids in the 8-neighborhood of the candidate grid mentioned in the detection process specifically means that if the current grid is a candidate grid, then the eight grids of the upper, lower, left, right, upper right, lower left, and lower right of the grid are counted.
  • the grid is also the number Nneighbour of candidate grids.
  • step S206 the grid does not participate in subsequent processing.
  • step S204 When the feature information satisfies condition 1, implement step S204 and mark the grid as a candidate grid; when the feature information does not meet condition 1, implement step S206 for the grid that does not meet the condition, and do not participate in post-processing.
  • Step S207 judging whether the data satisfies condition 2.
  • FIG. 3 is a schematic diagram of a roadside distribution according to an embodiment of the present application.
  • the condition 2 means that the number of candidate grids around the current candidate grid cannot be too many or too few, that is, Min_neighbour ⁇ Nneighbour ⁇ Max_neighbour.
  • Min_neighbor and Max_neighbor follow certain principles.
  • Figure 3 is a schematic diagram of a roadside distribution according to an embodiment of the present application.
  • the blue curve is the state in which road boundaries exist in the grid.
  • the grid point is the current candidate grid
  • the orange grid is the grid that is also a candidate grid in the 8 neighborhood of the current candidate grid.
  • step S208 is implemented; if condition 2 is not met, step S206 is implemented.
  • Step S208 marking the grid as a secondary candidate grid.
  • Step S209 searching for merging candidate regions, and counting the number of grids in each region after merging.
  • the search algorithm is used to merge the candidate grids, and the number of grids in each area after merging is counted.
  • Searching and merging candidate areas and counting the number of candidate grids in the detection process specifically refers to randomly selecting a candidate grid as a seed point, creating a new set of candidate grids A, and then judging whether there are candidate grids in the neighborhood of 8 seed points , if there is, put it into set A, and then perform the same operation on the judged point until all the grids connected to the seed point are put into set A. Then reselect a candidate grid that has not been visited as a seed point, and perform the above operations until all candidate points are visited. At this point, multiple candidate raster sets of all candidate regions can be obtained. Then count the number of grids in each candidate area as Ncandidate. It should be noted that this search method is not specifically limited as this embodiment of the application, and users can also use other search methods to complete this clustering, which are also within the scope of protection of this embodiment of the application.
  • Step S210 judging whether the data satisfies condition 3.
  • the embodiment of the present application determines that it is a false detection, and does not perform subsequent processing as a road boundary.
  • step S211 If condition 3 is met, step S211 will be implemented; if condition 3 is not met, step S206 will be implemented, and the remaining grids will not participate in post-processing.
  • Step S211 mark as a real roadside boundary candidate area.
  • the grids satisfying condition 3 are marked as real roadside boundary candidate areas.
  • Step S212 selecting candidate points for each candidate area grid.
  • each grid selects a candidate point according to certain rules.
  • the selection of candidate points in each candidate grid described in the detection process specifically refers to all our detections so far.
  • the classification is performed on the grid.
  • the grid has a certain size and there are many points in the grid. Using all the points for subsequent curve fitting is not only unreasonable but also increases the amount of calculation. Therefore, we select the center of gravity of all points in the grid as candidate points, and then perform subsequent processing.
  • the center of gravity is calculated as follows:
  • Step S213 fitting each area candidate point to a road boundary curve.
  • Cubic curve fitting is performed on all candidate points in each candidate area, and the fitted curve is the curve of the road boundary, that is, the detection of the road boundary is completed.
  • the embodiment of the present application uses the cubic curve equation for fitting, and the fitting method adopts the random sampling consensus algorithm.
  • the final fitted curve is the road edge curve.
  • the fitting method is not specifically limited to the embodiments of the present application, and other curve fitting methods that can perform the same function, such as the least square method, should also be within the protection scope of the embodiments of the present application.
  • the road image of the structured road is scanned to obtain the laser radar point cloud data of the structured road; the laser radar point cloud data is rasterized to generate multiple grids, and the The grid feature information of the point cloud in each grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple target candidate grids; for multiple targets
  • the candidate grid performs area marking to obtain at least one roadside candidate area of the structured road; based on the area candidate points in each roadside candidate area, the road edge of the structured road is generated.
  • a lidar-based detection device for road boundaries is also provided. It should be noted that the lidar-based road boundary detection device can be used to implement the lidar-based road boundary detection method in Embodiment 1.
  • Fig. 4 is a schematic diagram of a lidar-based road boundary detection device according to an embodiment of the present application.
  • the lidar-based road boundary detection device 400 may include: an acquisition component 401 , a processing component 402 , a screening component 403 , a marking component 404 and a generating component 405 .
  • the acquiring component 401 is configured to scan the road image of the structured road, and acquire the lidar point cloud data of the structured road.
  • the processing component 402 is configured to perform rasterization processing on the lidar point cloud data, generate multiple rasters, and obtain raster feature information of the point cloud in each raster.
  • the screening component 403 is configured to perform at least one screening process on multiple grids based on the grid feature information of the point cloud in each grid to obtain multiple target candidate grids.
  • the marking component 404 is configured to perform area marking on a plurality of target candidate grids to obtain at least one roadside candidate area of the structured road.
  • the generating component 405 is configured to generate an overall error of the target model based on the detection error of the original model and the detection error of the target model.
  • the screening component 403 includes: a first screening component, configured to use at least one filtering condition to filter the grid feature information of the point cloud in each grid, wherein the filtering condition includes at least one of the following: A filter condition, wherein the filter factors included in the first filter condition include: the height of the highest point in the grid, the height of the lowest point in the grid and the height difference of the point cloud in the grid; the second filter condition, wherein the second The filter factors included in the filter condition include: whether the number of candidate grids around the current candidate grid is within the range of the filter threshold.
  • the first screening component includes: a first screening sub-component, configured to use one or more screening factors in the first screening condition to filter the grid feature information of the point cloud in each grid to obtain The first group of candidate grids, wherein the point cloud in the first group of candidate grids includes the following features: there are no false edges, low obstacles, and point cloud data of objects with height differences.
  • the first screening component includes: a second screening sub-component, configured to perform secondary screening on the candidate grids in the first group of candidate grids by using the second filtering condition after obtaining the first group of candidate grids, A second group of candidate grids is acquired, wherein the second group of candidate grids is the candidate grids in the first group of candidate grids that have undergone secondary grid marking.
  • a second screening sub-component configured to perform secondary screening on the candidate grids in the first group of candidate grids by using the second filtering condition after obtaining the first group of candidate grids, A second group of candidate grids is acquired, wherein the second group of candidate grids is the candidate grids in the first group of candidate grids that have undergone secondary grid marking.
  • the second screening subcomponent is configured to use the second filtering condition to perform secondary screening on the candidate grids in the first group of candidate grids to obtain the second group of candidate grids through the following steps: detecting the first group The number of candidate grids in the respective neighborhoods of each candidate grid in the candidate grid, where the candidate grid is the grid in the first group of candidate grids; if any candidate grid in the first group of candidate grids is adjacent to If the number of candidate grids in the domain is within the range of the screening threshold, the candidate grids within the range of the screening threshold are marked for secondary grids.
  • the marking component 404 includes: a first processing component, configured to use a search algorithm to merge the candidate grids in the second group of candidate grids, and count the number of candidate grids in each candidate region after merging; A second processing component, configured to perform area marking on the second group of candidate grids that have been merged, and obtain at least one roadside candidate area, where the roadside candidate area includes a plurality of target candidate grids.
  • the device further includes: a first marking component, configured to mark the candidate area when the number of candidate grids in any candidate area exceeds a target threshold.
  • a first marking component configured to mark the candidate area when the number of candidate grids in any candidate area exceeds a target threshold.
  • the generating component 404 includes: a first acquiring component, configured to acquire multiple grids in any roadside candidate area; a second acquiring component, configured to select candidate points from multiple grids, and acquire each Candidate points in the curb candidate area; the fitting component is set to perform curve fitting on all the candidate points in each curb candidate area, wherein the fitted curve is the road boundary of the structured road.
  • the device further includes: a preprocessing component configured to preprocess the lidar point cloud data after acquiring the lidar point cloud data of the structured road, wherein the preprocessing includes at least one of the following: Cloud undefined value filtering, noisy point filtering, and point cloud height filtering.
  • a preprocessing component configured to preprocess the lidar point cloud data after acquiring the lidar point cloud data of the structured road, wherein the preprocessing includes at least one of the following: Cloud undefined value filtering, noisy point filtering, and point cloud height filtering.
  • the laser radar point cloud data of the structured road is obtained by using the acquisition component to scan the road image of the structured road; by using the processing component to rasterize the laser radar point cloud data, a plurality of raster Grid, and obtain the grid feature information of the point cloud in each grid; by using the screening component based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple targets Candidate raster; through the marking component, perform area marking on multiple target candidate grids to obtain at least one roadside candidate area of the structured road; through the generation component, based on the area candidate points in each roadside candidate area, thereby generating The road boundary of the structured road.
  • a non-volatile storage medium wherein the non-volatile storage medium includes a stored program, wherein when the program is running, the non-volatile storage medium is controlled
  • the device executes the lidar-based road boundary detection method described in any one of the embodiments of the present application.
  • Each functional module provided in the embodiment of the present application may be run in a lidar-based road boundary detection method or a similar computing device, and may also be stored as a part of a non-volatile storage medium.
  • Fig. 5 is a schematic structural diagram of a non-volatile storage medium according to an embodiment of the present application.
  • a program product 50 according to an embodiment of the present application is described, on which a computer program is stored, and when the computer program is executed by a processor, the program code that implements the following steps:
  • Rasterize the lidar point cloud data generate multiple grids, and obtain the grid feature information of the point cloud in each grid;
  • At least one screening process is performed on the multiple grids to obtain multiple target candidate grids
  • a road boundary of the structured road is generated.
  • the program code that implements the following steps: use at least one filter condition to filter the grid feature information of the point cloud in each grid, wherein the filter condition includes at least one of the following: One: the first filter condition, wherein the filter factors included in the first filter condition include: the height of the highest point in the grid, the height of the lowest point in the grid, and the height of the point cloud in the grid Poor; the second filtering condition, wherein the filtering factors included in the second filtering condition include: whether the number of candidate grids around the current candidate grid is within the filtering threshold range.
  • the following steps are implemented: using one or more screening factors in the first screening condition to perform screening processing on the grid feature information of the point cloud in each grid, A first group of candidate grids is obtained, wherein the point cloud in the first group of candidate grids includes the following features: no false edges, low obstacles, and point cloud data of objects with height differences.
  • the program code when the computer program is executed by the processor, the program code to implement the following steps: use the second screening condition to perform secondary screening on the candidate grids in the first group of candidate grids to obtain the second group of candidate grids, wherein , the second group of candidate grids is the candidate grids that have undergone secondary grid marking in the first group of candidate grids.
  • the following steps are implemented: detecting the number of candidate grids in the respective neighborhoods of each candidate grid in the first group of candidate grids, wherein the candidate grid is the first The rasters in the group of candidate rasters; if the number of candidate rasters in the neighborhood of any candidate raster in the first group of candidate rasters is within the range of the screening threshold, the candidate rasters within the range of the screening threshold are subjected to secondary rasterization. grid mark.
  • the program code for implementing the following steps use a search algorithm to merge the candidate grids in the second group of candidate grids, and count the candidate grids in each candidate area after merging
  • the number of candidate grids that have been merged is marked as a region, and at least one roadside candidate region is obtained, wherein the roadside candidate region includes a plurality of target candidate grids.
  • the computer program is also executed by the processor to implement the program code of the following steps: obtain multiple grids in any roadside candidate area; select candidate points from multiple grids, and obtain each roadside candidate area Candidate points within; Curve fitting is performed on all candidate points within each roadside candidate area, where the fitted curve is the road boundary of the structured road.
  • the program code that implements the following steps: preprocessing the lidar point cloud data, wherein the preprocessing includes at least one of the following: point cloud undefined value filtering, noise point filtering and Point cloud height filtering.
  • the non-volatile storage medium may also be configured as program codes of various preferred or optional method steps provided by the lidar-based road boundary detection method.
  • Non-volatile storage media may include a data signal carrying readable program code in baseband or as part of a carrier wave traveling as a data signal. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • a non-volatile storage medium may send, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
  • the program code contained in the non-volatile storage medium can be transmitted by any appropriate medium, including but not limited to wireless, cable, optical cable, radio frequency, etc., or any suitable combination of the above.
  • FIG. 6 is a schematic structural diagram of a processor according to an embodiment of the present application. As shown in FIG. 6 , the processor 60 is set to To run the program, wherein the program executes the lidar-based road boundary detection method described in Embodiment 1 when running.
  • the processor 60 may execute an operation program of the lidar-based road boundary detection method.
  • the processor 60 may be configured to perform the following steps:
  • Rasterize the lidar point cloud data generate multiple grids, and obtain the grid feature information of the point cloud in each grid;
  • At least one screening process is performed on the multiple grids to obtain multiple target candidate grids
  • a road boundary of the structured road is generated.
  • the processor 60 may also be configured to perform the following step: filter the grid feature information of the point cloud in each grid by using at least one filtering condition, wherein the filtering condition includes at least one of the following: The first filter condition, wherein the filter factors included in the first filter condition include: the height of the highest point in the grid, the height of the lowest point in the grid, and the height difference of the point cloud in the grid; The second filtering condition, wherein the filtering factors included in the second filtering condition include: whether the number of candidate grids around the current candidate grid is within a screening threshold range.
  • the processor 60 may also be configured to perform the following steps: use one or more screening factors in the first screening condition to filter the grid feature information of the point cloud in each grid to obtain the second A set of candidate grids, wherein the point cloud in the first set of candidate grids includes the following features: point cloud data without false edges, low obstacles, and objects with height differences.
  • the processor 60 may also be configured to perform the following steps: secondarily filter the candidate grids in the first group of candidate grids by using the second screening condition, and obtain the second group of candidate grids, wherein the second group of candidate grids is The second group of candidate grids is the candidate grids in the first group of candidate grids that have undergone secondary grid marking.
  • the processor 60 may also be configured to perform the following step: detecting the number of candidate grids in the respective neighborhoods of each candidate grid in the first group of candidate grids, wherein the candidate grids are the first group of candidate grids A raster within a raster; if the number of candidate rasters within the neighborhood of any candidate raster in the first group of candidate rasters is within the range of the screening threshold, the candidate rasters within the range of the screening threshold will be marked for secondary rasters .
  • the processor 60 may also be configured to perform the following steps: use a search algorithm to merge candidate grids in the second group of candidate grids, and count the number of candidate grids in each candidate area after merging ; Perform area marking on the merged second group of candidate grids, and obtain at least one roadside candidate area, where the roadside candidate area includes multiple target candidate grids.
  • the processor 60 may also be configured to perform the following steps: obtain multiple grids in any one roadside candidate area; select candidate points from multiple grids, and obtain each roadside candidate area
  • Candidate points Curve fitting is performed on all candidate points in each roadside candidate area, where the fitted curve is the road boundary of the structured road.
  • the processor 60 may also be configured to perform the following steps: preprocessing the lidar point cloud data, wherein the preprocessing includes at least one of the following: point cloud undefined value filtering, noise point filtering and point cloud Highly filtered.
  • the above-mentioned processor 60 can execute various functional applications and data processing by running software programs and modules stored in the memory, that is, realize the above-mentioned road boundary detection method based on lidar.
  • the disclosed technical content can be realized in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the components may be a logical function division.
  • multiple components or components may be combined or may be Integrate into another system, or some features may be ignored, or not implemented.
  • the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, and components or indirect coupling or communication connection of components may be in electrical or other forms.
  • the components described as separate components may or may not be physically separated, and the components displayed as components may or may not be physical components, that is, they may be located in one place, or may be distributed to multiple components. Part or all of the components can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional component in each embodiment of the embodiment of the present application may be integrated into one processing component, or each component may physically exist separately, or two or more components may be integrated into one component.
  • the above-mentioned integrated components can be implemented in the form of hardware or in the form of software functional components.
  • the integrated components are realized in the form of software functional components and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions for enabling a computer device (which may be a personal computer, server or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disc, etc., which can store program codes. .
  • the solution provided by the embodiment of the present application can be applied to the road boundary detection process based on lidar, scan the road image of the structured road, obtain the lidar point cloud data of the structured road; perform rasterization processing on the lidar point cloud data , generate multiple grids, and obtain the grid feature information of the point cloud in each grid; based on the grid feature information of the point cloud in each grid, perform at least one screening process on multiple grids to obtain multiple The target candidate grid; performing area marking on multiple target candidate grids to obtain at least one roadside candidate area of the structured road; based on the area candidate points in each roadside candidate area, generating the road boundary of the structured road, thereby
  • the technical effect that the current road detection algorithm can be applied to the solid-state laser radar field is realized, and the technical problem that the current road detection algorithm cannot be applied to the solid-state laser radar field is solved.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Optical Radar Systems And Details Thereof (AREA)
  • Traffic Control Systems (AREA)
  • Image Processing (AREA)

Abstract

本申请公开了一种基于激光雷达的道路边界检测方法和装置。其中，该方法包括：扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界。

Description

基于激光雷达的道路边界检测方法和装置
本申请要求于2021年12月31日提交中国专利局、申请号为202111676309.9、发明名称为“基于激光雷达的道路边界检测方法和装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及车辆领域，具体而言，涉及一种基于激光雷达的道路边界检测方法和装置。
背景技术
目前，自动驾驶都离不开道路边界的检测，道路边界的检测是辅助驾驶中为人类提供安全行驶区域的核心功能组件，也是为全自动无人驾驶提供有效检测范围与可行驶区域的重要前置功能组件。
在相关技术中，道路边缘检测算法大都基于机械式激光雷达点云展开，但由于机械式激光雷达具有价格高、体积大、不便于量产等缺点，固态激光雷达应运而生，且在量产自动驾驶汽车的占据越来越高的地位。由于扫描方式不同，固态激光雷达与机械式激光雷达所形成的点云也有很大不同，从而存在道路检测算法均无法很好的适用于固态激光雷达场景的问题。
针对上述当前道路检测算法无法适用于固态激光雷达场景的问题，目前尚未提出有效的解决方案。
发明内容
本申请实施例提供了一种基于激光雷达的道路边界检测方法和装置，以至少解决当前道路检测算法无法适用于固态激光雷达场景的技术问题。
根据本申请实施例的一个方面，提供了一种基于激光雷达的道路边界检测方法，包括：扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿 候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界。
可选地，基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，包括：采用至少一种筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，其中，筛选条件包括如下至少之一：第一筛选条件，其中，所述第一筛选条件包含的筛选因素包括：所述栅格内最高点的高度、所述栅格内最低点的高度和所述栅格内点云的高度差；第二筛选条件，其中，第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
可选地，采用第一筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，包括：采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
可选地，在得到第一组候选栅格之后，方法还包括：采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，第二组候选栅格为第一组候选栅格中进行了二次栅格标记的候选栅格。
可选地，采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，包括：检测第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，候选栅格为第一组候选栅格内的栅格；如果第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于筛选阈值范围内，则将处于筛选阈值范围内的候选栅格进行二次栅格标记。
可选地，在获取第二组候选栅格之后，对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域，包括：采用搜索算法对第二组候选栅格中的候选栅格进行合并处理，并统计合并后每个候选区域内候选栅格的数量；对进行了合并处理的第二组候选栅格进行区域标记，并获取至少一个路沿候选区域，其中，路沿候选区域包括多个目标候选栅格。
可选地，如果任意一个候选区域内的候选栅格数量超过目标阈值，则将候选区域进行区域标记。
可选地，基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界，包括：获取任意一个路沿候选区域内的多个栅格；从多个栅格中选取候选点，获取每个路沿候选区域内的候选点；对每个路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为结构化道路的道路边界。
可选地，在获取结构化道路的激光雷达点云数据之后，方法还包括：对激光雷达 点云数据进行预处理，其中，预处理包括如下至少之一：点云未定义值过滤，噪声点过滤以及点云高度过滤。
根据本申请实施例的另一方面，还提供了一种基于激光雷达的道路边界的检测装置，包括：获取组件，设置为扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；处理组件，设置为对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；筛选组件，设置为基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；标记组件，设置为对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；生成组件，设置为基于原始模型的检测误差和目标模型的检测误差，生成目标模型的总体误差。
根据本申请实施例的另一方面，还提供了一种非易失性存储介质。该非易失性存储介质包括存储的程序，其中，在程序运行时控制非易失性存储介质所在设备执行本申请实施例的基于激光雷达的道路边界检测方法。
根据本申请实施例的另一方面，还提供了一种处理器。该处理器设置为运行程序，其中，程序运行时执行本申请实施例的基于激光雷达的道路边界的检测。
在本申请实施例中，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边。也就是说，本申请检测流程中的使用候选区域中所有候选点结合道路边界的实际情况，进行曲线拟合，最终拟合的曲线即为道路边缘曲线，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。
附图说明
此处所说明的附图用来提供对本申请实施例的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例的一种基于激光雷达的道路边界检测方法的流程图；
图2是根据本申请实施例的另一种基于激光雷达的道路边界检测方法的流程图；
图3是根据本申请实施例的一种路沿分布情况的示意图；
图4是根据本申请实施例的一种基于激光雷达的道路边界检测方法装置的示意图；
图5是根据本申请实施例的一种非易失性存储介质的结构示意图；
图6是根据本申请实施例的一种处理器的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
根据本申请实施例，提供了一种基于激光雷达的道路边界检测方法的实施例，需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
图1根据本申请实施例的一种基于激光雷达的道路边界检测方法的流程图，如图1所示的基于激光雷达的道路边界检测方法流程图，该方法包括如下步骤：
步骤S102，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据。
在本申请上述步骤S102提供的技术方案中，扫描结构化道路的道路图像，从而获取结构化道路的激光雷达点云数据。
在该实施例中，结构化道路可以为城市道路中比较正规的结构化的路沿，也可以对路边灌木丛或者是道路临时施工出现的水马、临时挡板等进行检测，但是，对于乡村路等不具有明显的、明确的道路边界的非结构化道路，则不属于检测目标。
可选地，激光雷达点云数据可以为激光雷达传感器通过测定传感器发射器与目标物体之间的传播距离，分析出物体表面的反射能量的大小，反射波谱的幅度、频率和相位等信息，从而呈现出自动驾驶车辆行驶过程中收集数据的三维点云数据，其中，激光雷达传感器可以为双目相机、三维扫描仪等。比如，扫描激光雷达传感器拍摄的图形，基于相机的内在参数得到激光雷达点云数据。
步骤S104，对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息。
在本申请上述步骤S104提供的技术方案中，在得到结构化道路的激光雷达点云数据之后，对激光雷达点云数据进行栅格化处理，生成多个栅格，在栅格化过程中统计栅格内点云的特征信息。
在该实施例中，栅格化处理可以为将向量图形格式表示的图形转换成栅格图形，可以为将点云在水平坐标轴的两个方向平均划分为多个栅格，划分过程中分别统计每个栅格的特征信息，其中，特征信息可以为栅格内点云高度差，栅格内点云的最大高度，栅格内点云的最小高度等。
可选地，栅格内点云高度差可以用Zdiff表示，栅格内点云的最大高度可以用Zgrid_max表示，栅格内点云的最小高度可以用Zgrid_min表示。
举例而言，将点云在xy方向平均划分为m*n个栅格，划分过程中分别统计每个栅格的三个特征信息，包括栅格内点云高度差，栅格内点云的最大高度，栅格内点云的最小高度。
步骤S106，基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格。
在本申请上述步骤S106提供的技术方案中，基于获取的每个栅格内点云的栅格特征信息，同时根据设定的筛选条件，对栅格进行一次筛选处理，从而得到目标候选栅格。
可选地，在该实施例中，筛选处理可以为根据设定的条件筛选掉不符合条件的栅格，筛选处理至少一次，其中，筛选条件可以为根据特征信息所设定的至少一条的条件，比如，设定栅格内点云的最大高度，当统计的特征信息大于设定栅格内点云的最大高度时，则过滤掉此栅格，从而得到目标候选栅格，或/和，设定栅格内点云的最小高度，当统计的特征信息小于设定栅格内点云的最小高度时，则过滤掉此栅格，从而得到目标候选栅格，或/和，设定栅格内点云的最大高度差，当统计的特征信息大于设定栅格内点云的最大高度差时，则过滤掉此栅格，从而得到目标候选栅格，或/和，设 定栅格内点云的最小高度差，当统计的特征信息小于设定栅格内点云的最小高度差时，则过滤掉此栅格，从而得到目标候选栅格。
可选地，目标候选栅格可以为对得到的结构化道路的激光雷达点云数据栅格化处理之后，根据设定的条件筛选掉不符合条件的栅格之后所得到的栅格。
步骤S108，对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域。
在本申请上述步骤S108提供的技术方案中，对多个目标候选栅格进行区域标记，即选取目标候选栅格内所有的重心作为候选点并进行区域标记，从而得到结构化道路的至少一个路沿候选区域。
步骤S110，基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界。
在本申请上述步骤S110提供的技术方案中，使用候选区域中所有候选点进行曲线拟合，结合道路边界的实际情况，最终拟合出来道路边缘曲线，生成结构化道路的道路边界。
本申请上述步骤S102至步骤S110，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边。也就是说，本申请检测流程中的使用候选区域中所有候选点结合道路边界的实际情况，进行曲线拟合，最终拟合的曲线即为道路边缘曲线，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。
下面对该实施例的上述方法进行进一步介绍。
作为一种可选的实施例方式，步骤S106，基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，该方法还包括：采用至少一种筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，其中，筛选条件包括如下至少之一：第一筛选条件，其中，第一筛选条件包含的筛选因素包括：栅格内最高点的高度、栅格内最低点的高度和栅格内点云的高度差；第二筛选条件，其中，第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
在该实施例中，对获取的激光雷达点云数据进行栅格化，生成多个栅格信息，同 时，统计每个栅格内点云的栅格特征信息，当特征信息满足第一筛选条件时，则标记栅格将其作为候选栅格，其他不满足第一筛选条件的栅格不进行标记，并且后期不再做处理，然后，判断标记的候选栅格是否满足第二筛选条件，对于满足第二筛选条件的栅格进行标记，得到目标候选栅格。
在该实施例中，第一筛选条件包含的筛选因素可以包括：栅格内最高点的高度、栅格内最低点的高度、栅格内点云的最大高度差和栅格内点云的最小高度差，其中，栅格内最高点的高度、栅格内最低点的高度和栅格内点云的高度差可以为经过多次测试得到的数值，由用户设定并输入系统中。第二筛选条件包含的筛选因素可以包括当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。对于经过第一筛选条件留下来的候选栅格，统计候选栅格8邻域内候选栅格的个数，第二筛选条件用于筛选掉当前候选栅格周围的候选栅格数量过多或多少的栅格，其中，设定测试候选栅格数量的最大值和最小值，从而过滤掉数量过多或多少的栅格候选栅格。
可选地，在经过第一筛选条件和第二筛选条件的筛选之后，得到目标候选栅格。
作为一种可选的实施例方式，采用第一筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，包括：采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
在该实施例中，采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一筛选条件可以为：栅格特征信息中的高度差不可以大于设定的栅格内点云的最大高度差；栅格特征信息中的高度差不可以小于设定的栅格内点云的最小高度差；栅格特征信息中的最高点的高度不可以小于栅格内最低点的高度；栅格特征信息中的最低点的高度不可以大于栅格内最高点的高度，经过第一筛选数据的筛选得到第一候选数据。
可选地，栅格内点云的最大高度差可以用max_threshold表示，栅格特征信息中的高度差不可以大于设定的栅格内点云的最大高度差，即，Zdiff<max_threshold，从而过滤掉高楼等假边缘。栅格内点云的最小高度差可以用min_threshold表示，栅格特征信息中的高度差不可以小于设定的栅格内点云的最小高度差，即，Zdiff>min_threshold，从而过滤掉地面以及地面的小突起等低矮障碍物。栅格内最低点的高度可以用min_height表示，栅格特征信息中的最高点的高度不可以小于栅格内最低点的高度，即，Zgrid_max>min_height，从而过滤掉一些特殊的低矮障碍物。栅格内最高点的高度可以用max_height表示，栅格特征信息中的最低点的高度不可以大于栅格内最高点的高度，从而过滤掉一些距离地面较远的或者悬空有高度差的物体。
举例而言，设定测试时栅格内点云的最大高度差为3.0米，当栅格特征信息中的高度差大于设定的栅格内点云的最大高度差，则过滤掉，并标记符合条件的栅格；由于国家规定路沿石的顺直高度应在10厘米以上，因此，设定测试使用的栅格内点云的最小高度差为0.10米，当栅格特征信息中的高度差大于设定的栅格内点云的最小高度差时，则过滤掉，并标记符合条件的栅格；设定测试时栅格内最低点的高度为0.05米，栅格特征信息中的最高点的高度小于栅格内最低点的高度时，则过滤掉，并标记符合条件的栅格；设定测试时栅格内最高点的高度为0.2米，当栅格特征信息中的最低点的高度大于栅格内最高点的高度时，则过滤掉，并标记符合条件的栅格。
可选地，标记栅格的方式可以为标记用颜色、图形等进行标记，此处不做限制。
作为一种可选的实施例方式，在得到第一组候选栅格之后，该方法还包括：采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，第二组候选栅格为第一组候选栅格中进行了二次栅格标记的候选栅格。
在该实施例中，得到第一组候选栅格并标记，采用第二筛选条件对第一组标记过的候选数据进行筛选并标记，选取存在两次候选标记的栅格作为第二组候选栅格，其中，标记可以为用颜色、图形等，此处不做限定。
作为一种可选的实施例方式，采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，包括：检测第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，候选栅格为第一组候选栅格内的栅格；如果第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于筛选阈值范围内，则将处于筛选阈值范围内的候选栅格进行二次栅格标记。
在该实施例中，采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格。
可选地，第二筛选条件包含的筛选因素可以包括当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。对经过第一筛选条件得到的第一组候选栅格，统计第一组候选栅格8邻域内候选栅格的个数，其中，候选领域可以为上下左右、左上、左下、右上、右下。比如，若当前选择栅格为候选栅格，则统计该栅格上下左右、左上、左下、右上、右下八个栅格中同样是候选栅格的个数，其中，候选栅格的个数可以用neighbour表示。
可选地，第二筛选条件用于筛选掉当前候选栅格周围的候选栅格数量过多或多少的栅格，即，候选栅格的最个数大于等于候选栅格的最小个数并且小于等于候选栅格的最大个数，其中，候选栅格的最大个数可以用Max_neighbour，候选栅格的最小个 数可以用Min_neighbour表示。
作为一种可选的实施例方式，步骤S108，在获取第二组候选栅格之后，对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域，包括：采用搜索算法对第二组候选栅格中的候选栅格进行合并，并统计合并后每个候选区域内候选栅格的数量；对进行了合并处理的第二组候选栅格进行区域标记，并获取至少一个路沿候选区域，其中，所述路沿候选区域包括多个目标候选栅格。
在该实施例中，采用搜索算法对第二组候选栅格中的候选栅格进行合并，并统计合并后每个候选区域内候选栅格的数量；对进行了合并处理的第二组候选栅格进行区域标记，并获取多个目标候选区域。
可选地，搜索算法可以为随机选取一个第一组候选栅格作为种子点，新建候选栅格集合用于存放栅格，判断种子点8领域内是否存在候选栅格，如果有则放入集合，直至将与种子点想通的所有栅格均放入集合之后，重新选择一个未被访问过的候选栅格作为种子点，再次新建候选栅格集合，重复上述操作，从而得到多个候选区域的多个候选栅格集。统计每个候选栅格集的数量，其中，候选栅格集的数量可以用Ncandidate表示。
需要说明，在该实施例中，搜索算法不做具体的限定，可以通过其他搜索算法完成统计候选栅格集的数量。
作为一种可选的实施例方式，如果任意一个候选区域内的候选栅格数量超过目标阈值，则将候选区域进行区域标记。
在该实施例中，目标阈值可以为系统设定的值，通过Min_candidate表示，因为正常情况下道路边界都是连续出现的，所以当候选区域内栅格数量过少时，则认定为误检，不作为道路边界进行后续处理。
距离而言，设定测试时目标阈值为15，则候选栅格集的数量要大于目标阈值，即Ncandidate>Min_candidate，当候选栅格集的数量小于目标阈值时，则不将该候选栅格集作为道路边界，并不进行后续处理。
作为一种可选的实施例方式，步骤S110，基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界，包括：获取任意一个路沿候选区域内的多个栅格；从多个栅格中选取候选点，获取每个路沿候选区域内的候选点；对每个路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为结构化道路的道路边界。
在该实施例中，获取任意一个路沿候选区域内的多个栅格；从多个栅格中选取候 选点，获取每个路沿候选区域内的候选点；对每个路沿候选区域内的所有候选点进行曲线拟合，从而得到结构化道路的道路边界。
可选地，从多个栅格中选取候选点可以为选取栅格每所有点的重心作为候选点，从而减小计算量，结合道路边界的实际情况，选择合适的拟合方式对使用的候选区域中所有的候选点进行曲线拟合，从而得到结构化道路的道路边界。
可选地，拟合方式为可以完成曲线拟合的拟合方式，如最小二乘法、随机采样一致性算法等，此处不做具体限定。
举例而言，从多个栅格中选取候选点可以为选取栅格每所有点的重心作为候选点，选用三次曲线方程进行拟合，拟合方式采用随机采样一致性算法，最终拟合的曲线即为道路边缘曲线。
作为一种可选的实施例方式，在获取结构化道路的激光雷达点云数据之后，该方法还包括：对激光雷达点云数据进行预处理，其中，预处理包括如下至少之一：点云未定义的值过滤，噪声点过滤以及点云高度过滤。
在该实施例中，对激光雷达点云数据进行预处理可以为主要包括：点云未定义的值过滤，噪声点过滤以及点云高度过滤。其中，点云未定义的值过滤用于过滤掉原始点云数据中点云坐标异常的点，可以为通过遍历点云、判断点云坐标完成；噪声点过滤可以为通过点云栅格化完成，将点云从三个维度划分为两个维度的栅格，统计栅格内点云的数量，用户根据实际情况设定一个阈值，若单个栅格内点云少于用户设定的阈值，则认为栅格内点云均为噪声点，从而过滤掉噪声点；点云高度过滤可以为过滤掉点云绝对高度过高或者过低的点云。
该实施例在自动驾驶车辆行驶过程中，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边。也就是说，本申请检测流程中的使用候选区域中所有候选点结合道路边界的实际情况，进行曲线拟合，最终拟合的曲线即为道路边缘曲线，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。
下面结合优选的实施方式对本申请的技术方案进行举例说明。
随着科技的不断进步，自动驾驶为解放人类双手提供了更多的可能。纵观自动驾 驶的发展历程，无论是L2、L3等辅助驾驶还是L4、L5等更高级的自动驾驶都离不开道路边界的检测，它是辅助驾驶中为人类提供安全行驶区域的核心功能组件，也是为全自动无人驾驶提供有效检测范围与可行驶区域的重要前置功能组件。
以往的道路边缘检测算法大都基于机械式激光雷达点云展开，但由于机械式激光雷达具有价格高、体积大、不便于量产等缺点，固态激光雷达应运而生，且在量产自动驾驶汽车的占据越来越多的地位，但是，由于扫描方式不同，固态激光雷达与机械式激光雷达所形成的点云也有很大不同，所以之前道路检测算法均无法很好的适用于固态激光雷达场景下。
目前，现有道路边界检测算法主要存在以下几点局限性：(1)检测算法过于依赖机械式激光的扫描模型，即理想情况下雷达高度h、扫描线俯仰角α以及探测距离l满足h＝cos(α)*l，基于这个基本模型就可以对扫描物状态进行估计，进而进行道路边界检测，但是，对于固态激光雷达来说不满足这个基本物理模型，因而导致以往算法的准确性大大降低；(2)检测算法依赖额外的激光雷达线束信息，对于同一条扫描线来说，从地面扫描到路沿的过程会产生较大的高度及曲率突变，大部分道路边界检测算法都基于这个基本思想，该方法的前提就是需要获取同一条扫描线的点云，即需要雷达的线束信息，但对于一帧点云来说点的坐标，强度等才是基本信息，而激光线束是额外的。同时，固态激光雷达并不是一条扫描线扇形扫描，也大都不具备所谓的线束信息，因而导致基于线束信息的检测方法无法适用；(3)检测算法依赖结构化道路模型，检测对象比较单一。以往检测算法大都基于非常理想化的结构化道路模型，检测对象大都是城市道路中比较正规的结构化的路沿，而对于路边灌木丛或者是道路临时施工出现的水马、临时挡板等都无法进行检测。
在该实施例中，为了克服以上问题，在一种相关技术中，设计出了一种基于三维激光雷达的道路边界检测方法，通过使用栅格点云高度差做障碍物与非障碍物的判别，同时，对距离灰度图进行分析，并采用区域生长法获取区域轮廓，但该算法对于路边灌木丛或者是道路临时施工出现的水马、临时挡板等，仍存在无法进行检测的问题。
在另一种相关技术中，提出一种道路边缘检测方法和装置，该方法使用点云的法线和法向曲率进行道路边缘候选点提取，然后采用区域生长和凹包算法进行噪声点去除，但该算法同样对于路边灌木丛或者是道路临时施工出现的水马、临时挡板等，仍存在无法进行检测的问题。
在另一种相关技术中，提出一种基于点云局部凹凸特征的道路边界实时检测方法，该方法将点云转换为深度展开图像，然后利用图像的凹凸特征进行边界检测，从而实现道路检测，但该方法存在不适用于固态激光雷达点云的问题。
在另一种相关技术中，提出一种道路边界实时提取及测量方法和装置，该方法针对机械式激光雷达的扫描特点，将点云角度突变的点作为边界候选点，然后对左右两侧分别构建数学描述模型，从而提高道路检测过程中算法运行的速度，但是该方法存在不适用于固态激光雷达点云的问题。
为克服现有技术的局限性，在该实施例中，提出一种基于激光雷达的道路边界检测方法，该方法不需要借助额外的线束信息，仅凭借单帧点云中点的位置信息就可以进行道路边界的检测；不仅可以检测结构化道路中的正规的路沿，也可以检测路边灌木丛形成的道路边界以及由于道路施工放置的水马、临时挡板等形成的临时道路边界，且本申请实施例算法实时性较高，设计出了既适用于机械式激光雷达点云，也适用于固态激光雷达点云的检测模型。
该实施例的算法流程图如图2所示，图2是根据本申请实施例的一种基于激光雷达的道路边界检测方法的流程图，该方法可以包括以下步骤：
步骤S201，对点云进行预处理。
首先对点云进行预处理，对于检测流程中所述点云预处理主要包括点云不确定值过滤，噪声点过滤以及点云高度过滤三个部分。其中，不确定值过滤是指过滤掉原始点云中点云坐标异常的点，通过遍历点云，判断点云坐标即可完成；噪声点过滤通过点云栅格化完成，即将点云从xyz三个维度划分为m*n*l个栅格，统计栅格内点云的数量，若单个栅格内点云少于设定的阈值N(N可根据用户实际情况自行设定，本申请实施例测试采用N＝10，需说明，本申请实施例所有的参数设定仅作为参考与测试使用，不作为本申请实施例的具体限定)则认为栅格内点云均为噪声点；点云高度过滤只指过滤掉点云绝对高度过高或者过低的点云，因为对于道路边缘来说其高度不会过高或者过低，因此若点云高度方向坐标z满足z<Zmin或者z>Zmax(Zmin与Zmax的设定需要结合用户实际情况，即雷达安装高度、点云坐标系原点定位来决定，本申请实施例测试采用值为Zmin＝-2.0m，Zmax＝5.0m)，则过滤掉该点，不作为算法的感兴趣区域。
步骤S202，将点云栅格格式化，并统计栅格特征信息。
对点云进行预处理之后，对处理后的点进行栅格化，同时，栅格化过程中统计栅格内点云的特征信息。
检测流程中所述点云栅格化，并统计栅格内点云信息具体是指将点云在xy方向平均划分为m*n个栅格，划分过程中分别统计每个栅格的三个特征信息，包括栅格内点云高度差Zdiff，栅格内点云的最大高度Zgrid_max，栅格内点云的最小高度Zgrid_min。
步骤S203，判断数据是否满足条件1。
判断数据是否满足条件1，其中，检测流程中所述条件1具体指：1)Zdiff<max_threshold，即栅格内点云高度差不可过大，以此过滤掉高楼等假边缘，本申请实施例测试使用max_threshold＝3.0m；2)Zdiff>min_threshold，即栅格内点云的高度差不可过小，以此过滤掉地面以及地面的小突起等低矮障碍物，调查发现国家规定路沿石的顺直高度应在10cm以上，因此本申请实施例测试使用min_threshold＝0.10m；3)Zgrid_max>min_height，即栅格内最高点的高度不能过低，以过滤一些特殊的低矮障碍物，本申请实施例测试使用min_height＝0.05m；4)Zgrid_min<max_height，即栅格内点云的最低点不能过高，以此过滤掉一些距离地面较远的或者悬空的有高度差的物体，本申请实施例测试使用max_height＝0.2m。
步骤S204，把栅格标记为候选栅格。
对符合条件1的候选栅格进行标记为候选栅格。
步骤S205，统计候选栅格8领域候选栅格的数量。
对于候选栅格，计算每个栅格8邻域内候选栅格的数量。
检测流程中所述统计候选栅格8邻域内候选栅格的个数具体是指若当前栅格为候选栅格，那么就统计该栅格上下左右，左上，右上，左下，右下八个栅格中同样是候选栅格的个数Nneighbour。
步骤S206，栅格不参与后续处理。
当特征信息满足条件1时，则实施步骤S204，将栅格标记为候选栅格，当特征信息不满足条件1时，对于不满足条件的栅格实施步骤S206，不参与后期处理。
步骤S207，判断数据是否满足条件2。
判断候选上是否满足条件2，其中，所述条件2是指当前候选栅格周围的候选栅格数量不能过多或者过少，即Min_neighbour≤Nneighbour≤Max_neighbour。其中两个阈值参数Min_neighbour与Max_neighbour的设定遵循一定原则。如图3所示，图3是根据本申请实施例的一种路沿分布情况示意图，道路边界存在的状态一般有4种，蓝色的曲线是道路边界在栅格中存在的状态，蓝色的栅格点是当前候选栅格，橙黄色的栅格是当前候选栅格8邻域中同样是候选栅格的栅格，根据这个原则本申请实施例使用阈值为Min_neighbour＝2，Max_neighbour＝5。
满足条件2，则实施步骤S208；不满足条件2，则实施步骤S206。
步骤S208，将栅格标记为二次候选栅格。
对满足条件2的栅格标记为二次候选栅格。
步骤S209，搜索合并候选区域，并统计合并后每个区域的栅格数。
对于二次筛选后的候选栅格，采用搜索算法进行候选栅格的合并，统计合并后每个区域栅格的数量。
检测流程中所述的搜索合并候选区域并统计区域内候选栅格数量具体指，随机选取一个候选栅格作为种子点，新建候选栅格集合A，然后判断种子点8邻域内是否存在候选栅格，如果有则将其放入集合A，然后对判断该点执行同样上述操作，直至与种子点相通的所有栅格均被放入集合A。然后重新选择一个未被访问过的候选栅格作为种子点，执行上述操作，直至所有候选点均被访问。此时就可以得到所有候选区域的多个候选栅格集合。然后统计每个候选区域内栅格数量为Ncandidate。需说明，此搜索方法不作为本申请实施例具体限定，用户也可以使用其他搜索方法完成本次聚类，同样在本申请实施例的保护范围内。
步骤S210，判断数据是否满足条件3。
对候选栅格进行判断，其中，所述的条件3具体指Ncandidate>Min_candidate，即每个候选区域的候选栅格数应该大于设定的阈值，因为正常情况下道路边界都是连续出现的，所以当候选区域内栅格数量过少时本申请实施例认定为误检，不作为道路边界进行后续处理。本申请实施例设定Min_candidate＝15。
满足条件3，则实施步骤S211；不满足条件3，则实施步骤S206，其余栅格不参与后期处理。
步骤S211，标记为真正的路沿边界候选区域。
对满足条件3的栅格标记为真正的路沿边界候选区域。
步骤S212，对每个候选区域栅格选取候选点。
对于所有候选区域内的所有栅格，每个栅格按照一定规则选取一个候选点，其中，检测流程中所述的在每个候选栅格内选取候选点具体是指，截止目前我们所有的检测分类都是针对栅格进行的，栅格具有一定大小，栅格内也有很多的点，使用所有的点进行后续曲线拟合不仅不够合理也会增大计算量。所以我们选取栅格内所有点的重心作为候选点，再进行后续处理。重心计算方式如下：
Figure PCTCN2022118213-appb-000001
其中(x,y,z)为最终重心坐标，(xi,yi,zi)为候选栅格中第i个点的坐标，Ngrid为栅格内的点数。
步骤S213，将每个区域候选点拟合为道路边界曲线。
对每个候选区域内的所有候选点进行三次曲线拟合，拟合后的曲线即为道路边界的曲线，即完成道路边界的检测。
检测流程中所述的使用候选区域中所有候选点进行曲线拟合，结合道路边界的实际情况，本申请实施例选用三次曲线方程进行拟合，拟合方式采用随机采样一致性算法。最终拟合的曲线即为道路边缘曲线。同时，拟合方法同样不作为本申请实施例的具体限定，其他可以完成同样功能的曲线拟合方式如最小二乘法等也应在本申请实施例的保护范围内。
该实施例在自动驾驶车辆行驶过程中，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边。也就是说，本申请检测流程中的使用候选区域中所有候选点结合道路边界的实际情况，进行拟合，最终拟合的曲线即为道路边缘曲线，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。
根据本申请，还提供了一种基于激光雷达的道路边界的检测装置。需要说明的是，该基于激光雷达的道路边界的检测装置可以用于执行实施例1中的基于激光雷达的道路边界的检测方法。
图4是根据本申请实施例的一种基于激光雷达的道路边界的检测装置示意图。 如图4所示，该基于激光雷达的道路边界的检测装置400可以包括：获取组件401、处理组件402、筛选组件403、标记组件404和生成组件405。
获取组件401，设置为扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据。
处理组件402，设置为对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息。
筛选组件403，设置为基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格。
标记组件404，设置为对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域。
生成组件405，设置为基于原始模型的检测误差和目标模型的检测误差，生成目标模型的总体误差。
可选地，筛选组件403包括：第一筛选组件，设置为采用至少一种筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，其中，筛选条件包括如下至少之一：第一筛选条件，其中，第一筛选条件包含的筛选因素包括：栅格内最高点的高度、栅格内最低点的高度和栅格内点云的高度差；第二筛选条件，其中，第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
可选地，第一筛选组件包括：第一筛选子组件，设置为采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
可选地，第一筛选组件包括：第二筛选子组件，设置为在得到第一组候选栅格之后，采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，第二组候选栅格为第一组候选栅格中进行了二次栅格标记的候选栅格。
可选地，第二筛选子组件，设置为通过以下步骤来采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格：检测第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，候选栅格为第一组候选栅格内的栅格；如果第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于筛选阈值范围内，则将处于筛选阈值范围内的候选栅格进行二次栅格标记。
可选地，标记组件404包括：第一处理组件，设置为采用搜索算法对第二组候选 栅格中的候选栅格进行合并，并统计合并后每个候选区域内候选栅格的数量；第二处理组件，设置为对进行了合并处理的所述第二组候选栅格进行区域标记，并获取至少一个路沿候选区域，其中，路沿候选区域包括多个目标候选栅格。
可选地，所述装置还包括：第一标记组件，设置为在任意一个候选区域内的候选栅格数量超过目标阈值的情况下，将候选区域进行区域标记。
可选地，生成组件404包括：第一获取组件，设置为获取任意一个路沿候选区域内的多个栅格；第二获取组件，设置为从多个栅格中选取候选点，获取每个路沿候选区域内的候选点；拟合组件，设置为对每个路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为结构化道路的道路边界。
可选地，所述装置还包括：预处理组件，设置为在获取结构化道路的激光雷达点云数据之后，对激光雷达点云数据进行预处理，其中，预处理包括如下至少之一：点云未定义的值过滤，噪声点过滤以及点云高度过滤。
在本申请实施例中，通过使用获取组件扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；通过使用处理组件对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；通过使用筛选组件基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；通过标记组件，对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；通过生成组件，基于每个路沿候选区域中的区域候选点，从而生成结构化道路的道路边界。也就是说，本申请检测流程中的使用候选区域中所有候选点结合道路边界的实际情况，进行曲线拟合，最终拟合的曲线即为道路边缘曲线，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。
根据本申请实施例，还提供了一种非易失性存储介质，其中，该非易失性存储介质包括存储的程序，其中，在所述程序运行时控制所述非易失性存储介质所在设备执行本申请实施例中任意一项所述的基于激光雷达的道路边界检测方法。
本申请实施例所提供的各个功能模块可以在基于激光雷达的道路边界检测方法或者类似的运算装置中运行，也可以作为非易失性存储介质的一部分进行存储。
图5是根据本申请实施例的一种非易失性存储介质的结构示意图。如图5所示，描述了根据本申请的实施方式的程序产品50，其上存储有计算机程序，计算机程序被处理器执行时实现如下步骤的程序代码：
扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；
对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；
基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；
对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；
基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：采用至少一种筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，其中，筛选条件包括如下至少之一：第一筛选条件，其中，所述第一筛选条件包含的筛选因素包括：所述栅格内最高点的高度、所述栅格内最低点的高度和所述栅格内点云的高度差；第二筛选条件，其中，第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，第二组候选栅格为第一组候选栅格中进行了二次栅格标记的候选栅格。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：检测第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，候选栅格为第一组候选栅格内的栅格；如果第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于筛选阈值范围内，则将处于筛选阈值范围内的候选栅格进行二次栅格标记。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：采用搜索算法对第二组候选栅格中的候选栅格进行合并处理，并统计合并后每个候选区域内候选栅格的数量；对进行了合并处理的第二组候选栅格进行区域标记，并获取至少一个路沿候选区域，其中，路沿候选区域包括多个目标候选栅格。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：获取任意一个路沿候选区域内的多个栅格；从多个栅格中选取候选点，获取每个路沿候选区域内的候选点；对每个路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为结构化道路的道路边界。
可选地，计算机程序还被处理器执行时实现如下步骤的程序代码：对激光雷达点 云数据进行预处理，其中，预处理包括如下至少之一：点云未定义值过滤，噪声点过滤以及点云高度过滤。
可选地，在本实施例中，非易失性存储介质还可以被设置为基于激光雷达的道路边界检测方法提供的各种优选地或可选的方法步骤的程序代码。
可选地，本实施例中的具体示例可以参考上述实施例中所描述的示例，本实施例在此不再赘述。
非易失性存储介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了可读程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。非易失性存储介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
非易失性存储介质中包含的程序代码可以用任何适当的介质传输，包括但不限于无线、有线、光缆、射频等等，或者上述的任意合适的组合。
根据本申请实施例，还提供了一种处理器，该处理器设置为运行程序，图6是根据本申请实施例的一种处理器的结构示意图，如图6所示，该处理器60设置为运行程序，其中，所述程序运行时执行实施例1中所述的基于激光雷达的道路边界检测方法。
在发明本实施例中，上述处理器60可以执行基于激光雷达的道路边界检测方法的运行程序。
可选地，在本实施例中，处理器60可以被设置为执行下述步骤：
扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；
对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；
基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；
对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；
基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界。
可选地，处理器60可以还被设置为执行下述步骤：采用至少一种筛选条件对每个栅格内点云的栅格特征信息进行筛选处理，其中，筛选条件包括如下至少之一：第一筛选条件，其中，所述第一筛选条件包含的筛选因素包括：所述栅格内最高点的高度、所述栅格内最低点的高度和所述栅格内点云的高度差；第二筛选条件，其中，第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
可选地，处理器60可以还被设置为执行下述步骤：采用第一筛选条件中的一个或多个筛选因素，对每个栅格内点云的栅格特征信息进行筛选处理，得到第一组候选栅格，其中，第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
可选地，处理器60可以还被设置为执行下述步骤：采用第二筛选条件对第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，第二组候选栅格为第一组候选栅格中进行了二次栅格标记的候选栅格。
可选地，处理器60可以还被设置为执行下述步骤：检测第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，候选栅格为第一组候选栅格内的栅格；如果第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于筛选阈值范围内，则将处于筛选阈值范围内的候选栅格进行二次栅格标记。
可选地，处理器60可以还被设置为执行下述步骤：采用搜索算法对第二组候选栅格中的候选栅格进行合并处理，并统计合并后每个候选区域内候选栅格的数量；对进行了合并处理的第二组候选栅格进行区域标记，并获取至少一个路沿候选区域，其中，路沿候选区域包括多个目标候选栅格。
可选地，处理器60可以还被设置为执行下述步骤：获取任意一个路沿候选区域内的多个栅格；从多个栅格中选取候选点，获取每个路沿候选区域内的候选点；对每个路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为结构化道路的道路边界。
可选地，处理器60可以还被设置为执行下述步骤：对激光雷达点云数据进行预处理，其中，预处理包括如下至少之一：点云未定义值过滤，噪声点过滤以及点云高度过滤。
上述处理器60可以通过运行存储在存储器内的软件程序以及模块，从而执行各种功能应用以及数据处理，即实现上述的基于激光雷达的道路边界检测方法。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
在本申请实施例的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的技术内容，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如所述组件的划分，可以为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个组件或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，组件或组 件的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的组件可以是或者也可以不是物理上分开的，作为组件显示的部件可以是或者也可以不是物理组件，即可以位于一个地方，或者也可以分布到多个组件上。可以根据实际的需要选择其中的部分或者全部组件来实现本实施例方案的目的。
另外，在本申请实施例各个实施例中的各功能组件可以集成在一个处理组件中，也可以是各个组件单独物理存在，也可以两个或两个以上组件集成在一个组件中。上述集成的组件既可以采用硬件的形式实现，也可以采用软件功能组件的形式实现。
所述集成的组件如果以软件功能组件的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本申请的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。
工业实用性
本申请实施例提供的方案可以应用于基于激光雷达的道路边界检测过程中，扫描结构化道路的道路图像，获取结构化道路的激光雷达点云数据；对激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；基于每个栅格内点云的栅格特征信息，对多个栅格进行至少一次筛选处理，得到多个目标候选栅格；对多个目标候选栅格进行区域标记，得到结构化道路的至少一个路沿候选区域；基于每个路沿候选区域中的区域候选点，生成结构化道路的道路边界，从而实现了当前道路检测算法可以适用于固态激光雷达场的技术效果，解决了当前道路检测算法无法适用于固态激光雷达场的技术问题。

Claims (10)

  1. 一种基于激光雷达的道路边界检测方法，所述方法包括：
    扫描结构化道路的道路图像，获取所述结构化道路的激光雷达点云数据；
    对所述激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；
    基于所述每个栅格内点云的栅格特征信息，对所述多个栅格进行至少一次筛选处理，得到多个目标候选栅格；
    对所述多个目标候选栅格进行区域标记，得到所述结构化道路的至少一个路沿候选区域；
    基于每个所述路沿候选区域中的区域候选点，生成所述结构化道路的道路边界。
  2. 根据权利要求1所述的方法，其中，基于所述每个栅格内点云的栅格特征信息，对所述多个栅格进行至少一次筛选处理，包括：
    采用至少一种筛选条件对所述每个栅格内点云的栅格特征信息进行所述筛选处理，其中，所述筛选条件包括如下至少之一：
    第一筛选条件，其中，所述第一筛选条件包含的筛选因素包括：所述栅格内最高点的高度、所述栅格内最低点的高度和所述栅格内点云的高度差；
    第二筛选条件，其中，所述第二筛选条件包含的筛选因素包括：当前候选栅格周围的候选栅格数量是否处于筛选阈值范围内。
  3. 根据权利要求2所述的方法，其中，采用所述第一筛选条件对所述每个栅格内点云的栅格特征信息进行所述筛选处理，包括：
    采用所述第一筛选条件中的一个或多个筛选因素，对所述每个栅格内点云的栅格特征信息进行所述筛选处理，得到第一组候选栅格，其中，所述第一组候选栅格内的点云包括如下特征：不存在假边缘、低矮障碍物、有高度差物体的点云数据。
  4. 根据权利要求3所述的方法，其中，在得到第一组候选栅格之后，所述方法还包括：
    采用所述第二筛选条件对所述第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，其中，所述第二组候选栅格为所述第一组候选栅格中进行了二次栅格标记的候选栅格。
  5. 根据权利要求4所述的方法，其中，采用所述第二筛选条件对所述第一组候选栅格中的候选栅格进行二次筛选，获取第二组候选栅格，包括：
    检测所述第一组候选栅格中每个候选栅格在各自邻域内候选栅格的数量，其中，所述候选栅格为所述第一组候选栅格内的栅格；
    如果所述第一组候选栅格中任意一个候选栅格邻域内候选栅格的数量处于所述筛选阈值范围内，则将处于所述筛选阈值范围内的所述候选栅格进行所述二次栅格标记。
  6. 根据权利要求4所述的方法，其中，在获取第二组候选栅格之后，对所述多个目标候选栅格进行区域标记，得到所述结构化道路的至少一个路沿候选区域，包括：
    采用搜索算法对所述第二组候选栅格中的候选栅格进行合并处理，并统计合并后每个候选区域内候选栅格的数量；
    对进行了所述合并处理的所述第二组候选栅格进行区域标记，并获取所述至少一个路沿候选区域，其中，所述路沿候选区域包括多个目标候选栅格。
  7. 根据权利要求6所述的方法，其中，如果任意一个候选区域内的候选栅格数量超过目标阈值，则将所述候选区域进行所述区域标记。
  8. 根据权利要求1-7中任意一项所述的方法，其中，基于每个所述路沿候选区域中的区域候选点，生成所述结构化道路的道路边界，包括：
    获取任意一个所述路沿候选区域内的多个栅格；
    从所述多个栅格中选取候选点，获取每个路沿候选区域内的候选点；
    对每个所述路沿候选区域内的所有候选点进行曲线拟合，其中，拟合后的曲线为所述结构化道路的道路边界。
  9. 根据权利要求1-7中任意一项所述的方法，其中，在获取所述结构化道路的激光雷达点云数据之后，所述方法还包括：
    对所述激光雷达点云数据进行预处理，其中，所述预处理包括如下至少之一：点云未定义的值过滤，噪声点过滤以及点云高度过滤。
  10. 一种基于激光雷达的道路边界的检测装置，应用于自动驾驶车辆，所述装置包括：
    获取组件，设置为扫描结构化道路的道路图像，获取所述结构化道路的激光雷达点云数据；
    处理组件，设置为对所述激光雷达点云数据进行栅格化处理，生成多个栅格，并获取每个栅格内点云的栅格特征信息；
    筛选组件，设置为基于所述每个栅格内点云的栅格特征信息，对所述多个栅格进行至少一次筛选处理，得到多个目标候选栅格；
    标记组件，设置为对所述多个目标候选栅格进行区域标记，得到所述结构化道路的至少一个路沿候选区域；
    生成组件，设置为基于每个所述路沿候选区域中的区域候选点，生成所述结构化道路的道路边界。
PCT/CN2022/118213 2021-12-31 2022-09-09 基于激光雷达的道路边界检测方法和装置 WO2023124231A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111676309.9 2021-12-31
CN202111676309.9A CN114359866A (zh) 2021-12-31 2021-12-31 基于激光雷达的道路边界检测方法和装置

Publications (1)

Publication Number Publication Date
WO2023124231A1 true WO2023124231A1 (zh) 2023-07-06

Family

ID=81105241

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/118213 WO2023124231A1 (zh) 2021-12-31 2022-09-09 基于激光雷达的道路边界检测方法和装置

Country Status (2)

Country Link
CN (1) CN114359866A (zh)
WO (1) WO2023124231A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117491983A (zh) * 2024-01-02 2024-02-02 上海几何伙伴智能驾驶有限公司 实现可通行区域边界获取及目标相对位置判别的方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114359866A (zh) * 2021-12-31 2022-04-15 中国第一汽车股份有限公司 基于激光雷达的道路边界检测方法和装置
CN116047537B (zh) * 2022-12-05 2023-12-26 北京中科东信科技有限公司 基于激光雷达的道路信息生成方法及系统
CN115861959A (zh) * 2023-02-24 2023-03-28 中国第一汽车股份有限公司 一种车道线识别方法、装置、电子设备及存储介质
CN115840227B (zh) * 2023-02-27 2023-07-04 福思(杭州)智能科技有限公司 道路边缘的检测方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127113A (zh) * 2016-06-15 2016-11-16 北京联合大学 一种基于三维激光雷达的道路车道线检测方法
CN110320504A (zh) * 2019-07-29 2019-10-11 浙江大学 一种基于激光雷达点云统计几何模型的非结构化道路检测方法
CN113009453A (zh) * 2020-03-20 2021-06-22 青岛慧拓智能机器有限公司 矿山路沿检测及建图方法及装置
WO2021168854A1 (zh) * 2020-02-29 2021-09-02 华为技术有限公司 可行驶区域检测的方法和装置
CN114359866A (zh) * 2021-12-31 2022-04-15 中国第一汽车股份有限公司 基于激光雷达的道路边界检测方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127113A (zh) * 2016-06-15 2016-11-16 北京联合大学 一种基于三维激光雷达的道路车道线检测方法
CN110320504A (zh) * 2019-07-29 2019-10-11 浙江大学 一种基于激光雷达点云统计几何模型的非结构化道路检测方法
WO2021168854A1 (zh) * 2020-02-29 2021-09-02 华为技术有限公司 可行驶区域检测的方法和装置
CN113009453A (zh) * 2020-03-20 2021-06-22 青岛慧拓智能机器有限公司 矿山路沿检测及建图方法及装置
CN114359866A (zh) * 2021-12-31 2022-04-15 中国第一汽车股份有限公司 基于激光雷达的道路边界检测方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117491983A (zh) * 2024-01-02 2024-02-02 上海几何伙伴智能驾驶有限公司 实现可通行区域边界获取及目标相对位置判别的方法
CN117491983B (zh) * 2024-01-02 2024-03-08 上海几何伙伴智能驾驶有限公司 实现可通行区域边界获取及目标相对位置判别的方法

Also Published As

Publication number Publication date
CN114359866A (zh) 2022-04-15

Similar Documents

Publication Publication Date Title
WO2023124231A1 (zh) 基于激光雷达的道路边界检测方法和装置
CN106650640B (zh) 一种基于激光雷达点云局部结构特征的负障碍物检测方法
CN109444911B (zh) 一种单目相机和激光雷达信息融合的无人艇水面目标检测识别与定位方法
CN112801022B (zh) 一种无人驾驶矿卡作业区道路边界快速检测更新的方法
CN102103202B (zh) 一种融合影像的机载激光雷达数据半监督分类方法
Xu et al. Multiple-entity based classification of airborne laser scanning data in urban areas
CN106204547B (zh) 从车载激光扫描点云中自动提取杆状地物空间位置的方法
Chang et al. Detection of pavement distresses using 3D laser scanning technology
KR101258668B1 (ko) 한반도 통합형 기상 레이더 품질 관리 시스템 및 그 방법
Höfle et al. Urban vegetation detection using high density full-waveform airborne lidar data-combination of object-based image and point cloud analysis
CN103605953A (zh) 基于滑窗搜索的车辆兴趣目标检测方法
CN106199557A (zh) 一种机载激光雷达数据植被提取方法
Jin et al. A point-based fully convolutional neural network for airborne lidar ground point filtering in forested environments
CN104484882A (zh) 一种基于机载LiDAR数据的城市区域电力线检测方法
CN111208533A (zh) 一种基于激光雷达的实时地面检测方法
Xu et al. Entities and features for classifcation of airborne laser scanning data in urban area
Yin et al. Spherical coordinates based methods of ground extraction and objects segmentation using 3-D LiDAR sensor
Hu et al. A fast and simple method of building detection from LiDAR data based on scan line analysis
CN111507341B (zh) 一种目标边界框的调整方法、装置、设备及存储介质
Liu et al. Background filtering and object detection with roadside LiDAR data
Zhou et al. Segmentation algorithm for 3D LiDAR point cloud based on region clustering
CN112488190A (zh) 一种基于深度学习的点云数据分类方法及系统
Zaletnyik et al. LIDAR waveform classification using self-organizing map
Arefi From LiDAR point clouds to 3D building models
JP2013008310A (ja) 三次元地図作成装置およびその窓領域検出装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22913500

Country of ref document: EP

Kind code of ref document: A1