High-precision map drawing method

Abstract

The invention relates to a high-precision map drawing method, in particular to the technical field of intelligent driving. The method comprises the steps that in the vehicle driving process, point cloud data are collected through a laser radar, and each frame of point cloud data and corresponding vehicle pose information are recorded; and dividing the point cloud data into a straight line point cloud segment and a curve point cloud segment according to the vehicle pose information. And carrying out feature extraction on the linear point cloud segment, and determining a linear road edge and a linear lane line. And performing feature extraction on the curve point cloud segment, and determining a curve road edge and a curve lane line. And generating a high-precision map of the vehicle driving area according to the straight road edge, the curve road edge, the straight lane line and the curve lane line. The method has the beneficial effects of high road feature extraction precision, accurate road type identification, good map generation continuity and high map making efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, in particular to a high-precision map drawing method. BACKGROUND

[0002] At present, the high-precision map construction technology based on laser radar mainly relies on single geometric or reflectivity features to extract road edges and lane lines. In flat road scenes and the like, it can meet the basic requirements, but in complex road environments such as curved roads and intersections, due to the existence of environmental changes and various interference factors, the extraction accuracy and consistency often cannot meet the requirements of high-precision map construction.

[0003] Therefore, how to improve the extraction accuracy of road edges and lane lines in laser radar point cloud in complex road environments to meet the requirements of high-precision map construction has become a problem to be solved. SUMMARY

[0004] Therefore, it is necessary to provide a high-precision map drawing method to solve the above technical problems, which comprises:

[0005] In the process of vehicle driving, point cloud data is collected by laser radar, and each frame of point cloud data and corresponding vehicle pose information is recorded;

[0006] According to the vehicle pose information, the point cloud data is divided into straight line point cloud segments and curved road point cloud segments;

[0007] The straight line point cloud segments are subjected to feature extraction to determine straight line road edges and straight line lane lines;

[0008] The curved road point cloud segments are subjected to feature extraction to determine curved road edges and curved lane lines;

[0009] According to the straight line road edges, the curved road edges, the straight line lane lines and the curved lane lines, a high-precision map of the vehicle driving area is generated.

[0010] As an optional implementation manner, the laser radar is a mechanical rotating laser radar or a solid-state point array laser radar.

[0011] As an optional implementation manner, the step of dividing the point cloud data into straight line point cloud segments and curved road point cloud segments according to the vehicle pose information comprises:

[0012] The vehicle pose information is accumulated to calculate the cumulative rotation angle within a preset time length or a preset distance;

[0013] When the cumulative rotation angle is lower than a preset straight line maximum allowable angle threshold, the point cloud data corresponding to the vehicle pose information is determined as the straight line point cloud segment;

[0014] When the accumulated rotation angle is greater than a preset curve angle threshold, the point cloud data corresponding to the vehicle pose information is determined as the curve point cloud segment.

[0015] As an optional implementation, the method for determining the straight line alignment comprises:

[0016] The point cloud data of the straight line point cloud segment is divided into left and right sides based on the center axis of the vehicle;

[0017] For the point cloud data of each side, the straight line alignment points are identified based on the height jump and a preset angle threshold.

[0018] The straight line alignment points are fitted as the straight line alignment.

[0019] As an optional implementation, the method for identifying the straight line alignment points based on the height jump and the preset angle threshold for the point cloud data of each side comprises:

[0020] For each point in the point cloud data, the angle between the height difference of the point and its left and right nth points and the origin of the laser radar is calculated.

[0021] If the angle is greater than the angle threshold, the point is determined as a horizontal feature point.

[0022] For each horizontal feature point, the horizontal feature points with height jumps are screened as straight line alignment points in combination with the height variance.

[0023] As an optional implementation, the method for determining the straight line lane line comprises:

[0024] The point cloud data in the region between the straight line alignments is extracted.

[0025] In the point cloud data in the region between the straight line alignments, the point cloud with reflectivity greater than a preset first reflectivity threshold is determined as a straight line lane line point.

[0026] The straight line lane line points are fitted as the straight line lane line.

[0027] As an optional implementation, the method for performing feature extraction on the curve point cloud segment to determine the curve alignment and the curve lane line comprises:

[0028] Based on the vehicle pose information and a preset alignment width, a virtual alignment is expanded in the curve point cloud segment.

[0029] The point cloud located between the virtual alignments and having reflectivity greater than a preset second reflectivity threshold and point cloud density greater than a preset point cloud density threshold is determined as a candidate feature point.

[0030] perform curvature analysis on the point cloud of the candidate feature points to identify the curve lane edge points and the curve lane line points of the curve point cloud segment;

[0031] fit the curve lane edge points and the curve lane line points into the curve lane edge and the curve lane line respectively.

[0032] As an optional implementation, the curvature analysis on the point cloud of the candidate feature points to identify the curve lane edge points and the curve lane line points of the curve point cloud segment comprises:

[0033] cluster the candidate feature points;

[0034] for each cluster block, calculate the distance between the centroid position of the cluster block and the virtual lane edge and the point cloud curvature, and according to a preset classification rule, distinguish and aggregate the point cloud corresponding to the distance and the point cloud curvature into the curve lane edge points and the curve lane line points.

[0035] As an optional implementation, the generation of the high-precision map of the vehicle driving area according to the straight lane edge, the curve lane edge, the straight lane line and the curve lane line comprises:

[0036] splice the straight lane edge and the curve lane edge;

[0037] splice the straight lane line and the curve lane line;

[0038] generate continuous road lane edges and road lane lines through a preset fitting algorithm;

[0039] perform interval sampling and voxel filtering processing on the point cloud on the road lane edge and the road lane line, convert the point cloud into discrete points with consistent intervals, and generate the high-precision map of the vehicle driving area.

[0040] As an optional implementation, the method further comprises:

[0041] obtain local high-precision maps generated by multiple vehicles respectively in different vehicle driving areas;

[0042] perform layer splicing and error correction on the high-precision maps of the multiple vehicles through a preset adaptive splicing algorithm to obtain a global high-precision map of multiple vehicle driving areas.

[0043] In a second aspect, a computer device is provided, comprising a memory and a processor, the memory having stored thereon a computer program capable of running on the processor, and the processor, when executing the computer program, implements the method steps of any one of the first aspect.

[0044] In a third aspect, a computer readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, implements the method steps of any of the first aspect.

[0045] The application provides a high-precision mapping method, which comprises: collecting point cloud data by a laser radar during vehicle driving, and recording each frame of the point cloud data and corresponding vehicle pose information; dividing the point cloud data into straight line point cloud segments and curved road point cloud segments according to the vehicle pose information; performing feature extraction on the straight line point cloud segments to determine straight line road edges and straight line lane lines; performing feature extraction on the curved road point cloud segments to determine curved road edges and curved lane lines; and generating a high-precision map of the vehicle driving area according to the straight line road edges, the curved road edges, the straight line lane lines and the curved lane lines. The technical scheme provided by the embodiments of the application brings at least the following beneficial effects: by collecting point cloud data in real time by a laser radar during vehicle driving and synchronously recording the vehicle pose information corresponding to each frame of point cloud, the consistency of the point cloud data and the spatial motion state of the vehicle is ensured, which helps to improve the accuracy of subsequent spatial registration. In the point cloud data processing stage, according to the vehicle pose information, the point cloud data is automatically divided into straight line point cloud segments and curved road point cloud segments by accumulating the change in the rotation angle within a certain time or distance. By this segmentation method based on the change in the motion trajectory, the road types can be effectively distinguished, and the misjudgment problem caused by the traditional method based on static point cloud density or simple curvature estimation is avoided, and the accuracy of road type recognition is improved. For the straight line point cloud segments, the straight line road edge points are extracted by taking the vehicle center axis as the reference, combining the point cloud height jump characteristics and the horizontal direction angle judgment, and further screening the lane line points between the road edges based on the reflectivity threshold, so that the road boundaries and lane lines of the straight line road segments are accurately extracted. For the curved road point cloud segments, virtual road edges are constructed on the left and right sides of the vehicle, and the road edge points and lane line points in the curved road area are comprehensively screened and clustered by combining the point cloud reflectivity, density and curvature characteristics. By introducing curvature analysis, the curved road edges can be more accurately recognized, and different types of curved lane lines can be effectively distinguished, improving the reliability and accuracy of feature extraction in complex curved road conditions. In the map generation stage, the extracted straight line road edges, curved road edges, straight line lane lines and curved lane lines are respectively spliced and processed, and a fitting algorithm is used to generate continuous road edge lines and lane lines, and further through interval sampling and voxel filtering processing, the continuous line features are converted into regularly distributed discrete point clouds, ensuring that the generated high-precision map has high geometric precision, good data sparsity and processing efficiency. Therefore, by dynamically dividing the road types according to the vehicle pose information, extracting the road edge and lane line features based on the geometric characteristics and reflectivity characteristics, and generating a continuous road network through efficient data processing and fitting, the problems of difficult road feature extraction, large recognition error of straight line and curved road areas, discontinuity of the map and low overall generation efficiency in the background art are effectively solved, and the application has the beneficial effects of high road feature extraction accuracy, accurate road type recognition, good continuity of the generated map and high mapping efficiency.

[0046] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present application. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0048] Figure 1 A flow chart of a high-precision mapping method provided for the embodiments of the present application;

[0049] Figure 2 A flow chart of a point cloud segment division method provided for the embodiments of the present application;

[0050] Figure 3 A flow chart of a straight line road edge determination method provided for the embodiments of the present application;

[0051] Figure 4 A flow chart of a straight line road edge point identification method provided for the embodiments of the present application;

[0052] Figure 5 A flow chart of a straight line lane line determination method provided for the embodiments of the present application;

[0053] Figure 6 A flow chart of a curved road edge and curved lane line determination method provided for the embodiments of the present application;

[0054] Figure 7 A flow chart of a curvature analysis method provided for the embodiments of the present application;

[0055] Figure 8 A flow chart of a high-precision map generation method provided for the embodiments of the present application;

[0056] Figure 9 A structural schematic diagram of a computer device provided for the embodiments of the present application. DETAILED DESCRIPTION

[0057] In order to make the purposes, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0058] A high-precision mapping method provided by an embodiment of the present application will be described in detail below with reference to the specific embodiments. Figure 1 A flowchart of the high-precision mapping method provided by an embodiment of the present application is shown in FIG. 1. Figure 1 The specific steps are as follows:

[0059] In step 101, point cloud data is collected by a laser radar during vehicle driving, and each frame of point cloud data and corresponding vehicle pose information are recorded.

[0060] In implementation, the laser radar scans the surrounding environment through laser beams to form dense three-dimensional point cloud data. Each point in the point cloud can contain position information (X, Y, and Z) and optional reflectivity information. The vehicle pose information (including position and attitude, i.e., translation and rotation) records the accurate state of the vehicle in the global or local coordinate system when each frame of point cloud is generated. Synchronous collection of point cloud data and pose information can achieve spatial registration and trajectory accumulation of subsequent point cloud data. For example, a mechanical rotating laser radar can be installed on the top of the vehicle, and point cloud data at a frequency of 10 Hz per second can be collected. The position information (which can include latitude, longitude, and elevation) and the orientation angle (heading angle, pitch angle, and roll angle) corresponding to each frame of point cloud data can be recorded by a vehicle-mounted inertial navigation system (INS) and GNSS combined navigation.

[0061] As an optional implementation, the laser radar is a mechanical rotating laser radar or a solid-state dot array laser radar.

[0062] In step 102, the point cloud data is divided into straight line point cloud segments and curved point cloud segments according to the vehicle pose information.

[0063] In implementation, the heading change (i.e., change in rotation angle) in the vehicle pose can reflect the motion trend of the vehicle on the road. By accumulating the heading change angle of the vehicle within a certain time or distance, it can be determined whether the current paragraph belongs to a straight line segment (small heading change) or a curved segment (large heading change). Dividing the straight line segment and the curved segment can help to apply different feature extraction algorithms in subsequent applications and improve the accuracy of feature recognition. For example, the vehicle driving path can be continuously sampled, and the cumulative rotation angle of the current path segment can be calculated every 5 meters of driving. If the cumulative rotation angle is less than 5°, the corresponding point cloud data is marked as a straight line point cloud segment, and if the cumulative rotation angle is greater than 15°, the corresponding point cloud data is marked as a curved point cloud segment.

[0064] As an optional implementation, Figure 2 A flowchart of the point cloud segment division method provided by an embodiment of the present application is shown in FIG. 2. Figure 2 The specific steps of dividing the point cloud data into straight line point cloud segments and curved point cloud segments according to the vehicle pose information in step 102 are as follows:

[0065] Step 201, accumulate vehicle pose information, calculate the cumulative rotation angle within a preset time length or a preset distance.

[0066] In implementation, by continuously obtaining pose data during vehicle driving, sampling according to time interval or driving distance, accumulating vehicle heading angle change in each section, the cumulative rotation angle of the vehicle in the section can be obtained. The cumulative rotation angle can reflect the bending degree of the vehicle motion path, and is a basic index for subsequent determination of straight line or curve. For example, the vehicle collects the heading angle every 1 meter, and the absolute value of the change of the collected heading angle within 10 meters is summed up as the cumulative rotation angle. For example, the total change of the heading angle within 10 meters is less than 5°, and the cumulative rotation angle is recorded as 5°. If the total change reaches 20°, the cumulative rotation angle is recorded as 20°.

[0067] Step 202, when the cumulative rotation angle is lower than the preset maximum allowable angle threshold of straight line, the point cloud data corresponding to the vehicle pose information is determined as a straight line point cloud section.

[0068] In implementation, if the cumulative rotation angle of the vehicle within a period of time or distance is small, it indicates that the vehicle basically drives along a straight path. A maximum allowable angle threshold of straight line is set as a judgment standard. When the cumulative rotation angle is lower than the threshold, it is determined that the point cloud corresponding to the current section is a straight line road section. For example, the maximum allowable angle threshold of straight line is set as 8°. In step 201, if the cumulative rotation angle within 10 meters is 5°, since 5° is less than 8°, the point cloud data of this section is marked as a straight line point cloud section.

[0069] Step 203, when the cumulative rotation angle is greater than the preset curve angle threshold, the point cloud data corresponding to the vehicle pose information is determined as a curve point cloud section.

[0070] In implementation, when the heading angle of the vehicle changes greatly within a period of time or distance, it indicates that the vehicle is passing through a curve area. A curve angle threshold is set as a judgment standard. When the cumulative rotation angle is greater than the threshold, it is determined that the corresponding point cloud is a curve section, so as to trigger the feature extraction method specially for curve scene. For example, the curve angle threshold is set as 15°. In step 201, if the cumulative rotation angle within 10 meters is 20°, since 20° is greater than 15°, the point cloud data of the corresponding interval is marked as a curve point cloud section for subsequent extraction and processing of curve road alignment and curve lane line.

[0071] Step 103, performing feature extraction on the straight line point cloud section to determine the straight line alignment and straight line lane line.

[0072] In implementation, the straight line road feature is relatively regular, and the road edge (road along) can be identified based on the height change and lateral angle characteristics of the point cloud, and the lane line can be extracted based on the reflectivity characteristics in the area between the two road alongs. Through side-by-side processing and feature screening, the road profile of the straight line road section can be effectively extracted. For example, the point cloud can be divided into left and right sides with the vehicle traveling direction as the central axis. For each side of the point cloud, the height difference and horizontal angle of adjacent points are calculated point by point, and the points with height mutation and angle exceeding 10° are screened as road along points. Then, the point cloud with reflectivity greater than 10% in the area between the left and right road alongs is extracted, and the straight lane line is obtained by straight line fitting.

[0073] As an optional implementation, Figure 3 A flowchart of a method for determining a straight line road along according to an embodiment of the present application is shown in Figure 3 As shown in step 103, the determination of the straight line road along is as follows:

[0074] Step 301: The point cloud data of the straight line point cloud section is divided into left and right sides with the central axis of the vehicle as the reference.

[0075] In implementation, the vehicle traveling direction corresponds to the central axis of the vehicle, which can be determined by the current pose information (such as the heading angle) of the vehicle. Therefore, the collected point cloud data can be divided into left and right point clouds according to the lateral coordinates with the central axis as the reference, which facilitates the subsequent extraction of the road boundary features of the left and right sides and improves the accuracy and processing efficiency of the extraction. For example, the central axis of the vehicle can be determined based on the current position and heading angle of the vehicle, and a local rectangular coordinate system is established. With the center of the vehicle as the origin, the forward direction as the X-axis, and the left and right directions as the Y-axis, the points with Y-axis coordinates less than 0 in the point cloud are divided into left point clouds, and the points with Y-axis coordinates greater than 0 are divided into right point clouds, and the left and right point cloud data are saved respectively.

[0076] Step 302: For each side of the point cloud data, the straight line road along points are identified based on the height jump and the preset angle threshold.

[0077] In implementation, the road edge shows obvious height jump, such as the road along stone being higher than the road surface. By analyzing the height difference between adjacent points for each side of the point cloud, and combining the angle change of the line connecting the point and the adjacent point with the horizontal plane of the laser radar, the points that meet the jump feature and the angle condition are screened as possible straight line road along points, so as to accurately identify the road edge. For example, the left and right point clouds can be sorted from near to far, and for each point, the height difference between the point and the fifth adjacent point on the left and right sides is calculated, and the angle between the connected straight line and the horizontal plane is calculated. If the height difference exceeds 0.1 meters and the angle is greater than 10°, the point is marked as a candidate straight line road along point. Further, the candidate point density is counted to screen the final straight line road along point.

[0078] As an optional implementation,Figure 4 A flowchart of a straight road edge point recognition method provided by an embodiment of the present application is shown in FIG. 3. As shown in FIG. 3, the specific steps of recognizing the straight road edge point of each side of the point cloud data based on the height jump and the preset angle threshold in step 302 are as follows: Figure 4

[0079] Step 401: For each point in the point cloud data, the height difference between the point and its left and right nth points and the angle of the laser radar origin are calculated.

[0080] In implementation, the road edge is accompanied by significant height changes in the continuously collected point cloud. In order to preliminarily screen the area where the road edge feature may exist, for each point, its left and right nth neighboring points are selected, and the height difference between these points is calculated to determine whether there is a significant height change. At the same time, in order to further accurately recognize the point with road edge characteristics, the horizontal angle of the laser radar origin to the line formed by the current point and its neighboring points is calculated. The calculation of the height difference is used for preliminary screening of the jump area, and the calculation of the angle change is used for accurate determination of the road edge characteristics, so as to improve the accuracy and robustness of feature extraction. For example: for each point cloud point P i , its left 5th point P i-5 and right 5th point P i+5 are selected, and the height differences Δh L = h i -h i-5 and Δh R = h i -h i+5 are calculated. Then, taking the laser radar installation point as the origin, the angle of the line from the origin to P i , P i-5 and P i+5 to the horizontal plane is calculated. For example, when a certain point P i collected during the vehicle driving process has a height of 1.2 meters, the left 5th point P i-5 has a height of 1.4 meters, and the right 5th point P i+5 has a height of 1.1 meters, the left and right height differences are calculated. At the same time, the angle of the line from the laser radar to these points is calculated as a judgment basis.

[0081] Step 402: If the angle is greater than the angle threshold, the point is determined as a horizontal feature point.

[0082] ​In practice, at road edges, due to sudden changes in terrain, the angle between the line connecting points and the horizontal plane is typically large. After setting an angle threshold, when the calculated angle exceeds this threshold, the current point is determined to have significant terrain change characteristics and is therefore a horizontal feature point. For example, the angle threshold is set to 10°. The angle calculated in step 401 is evaluated. If the angle between a point and its neighboring point is 12°, which is greater than 10°, the point is marked as a horizontal feature point. If the angle is 8°, the point is not marked.

[0083] In step 403, for each horizontal feature point, combined with the height variance, horizontal feature points with height jumps are screened out as straight roadside points.

[0084] During implementation, some non-edge areas of the road may have local noise or anomalies, so further screening based on the statistical characteristics of height changes is required. By calculating the height variance of the local area around the horizontal feature point, points with drastic height changes and that meet the jump characteristics are screened out, further ensuring that the final screened points belong to the straight roadside. For example: for each horizontal feature point, the height values ​​of the 10 points in its left and right neighborhoods are counted and the height variance is calculated. If the height variance is greater than 0.01 square meters, the horizontal feature point is determined to be a valid straight roadside point with a height jump. Otherwise, the point is eliminated and not processed as a roadside point.

[0085] Step 303: Fit the straight roadside points into a straight roadside.

[0086] In practice, curve or straight line fitting can be performed on the extracted discrete straight roadside points to form a continuous, regular road edge line. This fitting operation can smooth out local noise, improving the overall continuity and accuracy of the road boundary line, facilitating subsequent use in high-precision map generation. For example, the least squares method can be used to perform straight line fitting on the identified straight roadside points on the left and right sides, respectively, to obtain the left and right roadside lines. The fitting results are expressed as polynomial expressions or piecewise straight lines, and the fitting accuracy prioritizes ensuring the continuity of the edge line and the consistency of the actual road shape.

[0087] As an optional implementation, Figure 5 A flow chart of a method for determining a straight lane line provided in an embodiment of the present application is shown as follows: Figure 5 As shown, the steps for determining the straight lane line in step 103 are as follows:

[0088] Step 501: extract point cloud data of the area between straight road edges.

[0089] In practice, straight curbs usually mark the left and right boundaries of the road driving area. The point cloud data between the two curbs contains road marking information such as lane lines. By extracting the point cloud data between the left and right straight curbs, the area of ​​interest can be effectively limited and irrelevant environmental points outside the curbs can be removed, thus providing a clean data basis for accurate lane line extraction. For example: Assume that the fitting lines of the left straight curb and the right straight curb are obtained through the previous steps, and are denoted as L left and L right Extraction satisfies L left and L right All point cloud data between two straight curbs, with the longitudinal distance (Y-axis coordinate) within the range between the two lines. For example, if the left curb is at Y = -3 meters and the right curb is at Y = 3 meters, then point clouds with Y coordinates in the range [-3 meters, 3 meters] are selected.

[0090] Step 502 : In the point cloud data of the area between the straight roadside edges, point clouds having a reflectivity greater than a preset first reflectivity threshold are determined as straight lane line points.

[0091] In practice, lane lines on the road surface are usually made of highly reflective materials, which have a high reflectivity in the laser radar echo. By setting a reflectivity threshold and screening out point clouds with reflectivity higher than the threshold, the points scattered on the lane lines can be effectively identified, thereby distinguishing them from ordinary road surface points. For example: set the first reflectivity threshold to 10%. Traverse the point cloud data extracted in step 501 and retain all points with a reflectivity greater than 10% as straight lane line points. For example, if the reflectivity of a point cloud is 12%, it will be retained if it is greater than 10%. If the reflectivity is 8%, it will be eliminated and not processed as a lane line point.

[0092] Step 503: Fit the straight lane line points into a straight lane line.

[0093] In practice, the selected lane points are typically distributed in a spatial strip along the vehicle's travel direction. To eliminate the effects of local noise and improve continuity, a fitting algorithm can be used to fit the discrete lane points to generate a continuous lane model that conforms to the road geometry. For example, the lane points obtained in step 502 can be processed using a least squares linear fitting method to produce a straight line equation. During the fitting process, algorithms such as RANSAC can be used for robustness optimization to eliminate the influence of outliers and ensure good accuracy and stability of the fitting results.

[0094] Step 104 : extract features from the curve point cloud segment to determine the curve curb and curve lane line.

[0095] In implementation, the road features of the curved section are complex, and it is difficult to accurately extract based on the point cloud density and reflectivity directly. By constructing a virtual road edge frame to define the extraction range, and combining with the local curvature analysis of the point cloud, the road boundary and lane line position in the curved section can be effectively identified, solving the problem of large feature extraction error in the curved section. For example: based on the vehicle pose, a virtual road edge area can be set 5 meters outward on the left and right sides of the vehicle, and the point cloud located between the left and right virtual road edges is screened. The points with reflectivity higher than 8% and point cloud density greater than the preset point cloud density threshold are clustered, and the curvature of each cluster block is calculated. According to the distance from the centroid to the virtual road edge and the curvature size, the cluster blocks are classified as curved road edge points or curved lane line points, and curve fitting is performed respectively.

[0096] As an optional implementation, Figure 6 A flowchart of a method for determining a curved road edge and a curved lane line provided by an embodiment of the present application is shown in Figure 6 As shown in step 104, the specific steps of determining the curved road edge and the curved lane line by extracting features from the curved point cloud section are as follows:

[0097] Step 601: Based on the vehicle pose information and the preset road edge width, a virtual road edge is expanded in the curved point cloud section.

[0098] In implementation, in the curved section, it is difficult to directly rely on the point cloud geometric features to define the road boundary due to the continuous change of the vehicle heading. By expanding the virtual road edge area with a certain width on the left and right sides based on the current position and orientation of the vehicle, the approximate range of the actual road boundary can be estimated, thereby providing a stable and limited spatial range for subsequent feature point screening in the dynamically changing curved environment. For example: based on the current pose of the vehicle as a reference, a virtual road edge belt can be constructed by expanding outward 3 meters on the left and right sides of the vehicle. Assuming that the vehicle driving direction is the positive direction of the X axis, then in the transverse (Y axis) direction of the vehicle, the positive and negative directions are respectively translated by 3 meters, forming the left and right boundary areas to limit the extraction range of the curved point cloud.

[0099] Step 602: The point cloud located between the virtual road edges and having reflectivity greater than a preset second reflectivity threshold and point cloud density greater than a preset point cloud density threshold is determined as a candidate feature point.

[0100] In implementation, the point cloud with reflection characteristics and distribution density is screened in the area defined by the virtual road edge. By setting the reflectivity threshold and the point cloud density threshold, discrete noise points and irrelevant road points can be effectively removed, and only candidate feature points with potential road boundary or lane line characteristics are retained, for example, the second reflectivity threshold is set to 8%, and the point cloud density threshold is set to 0.5 points per square meter. In the screening step 601, all points with a reflectivity greater than 8% in the area between the virtual road edges are screened, and then the density statistics are performed based on the number of neighbor points within a local radius of 0.5 meters, and the points with a density greater than 0.5 points per square meter are retained as candidate feature points.

[0101] In step 603, curvature analysis is performed on the point cloud of the candidate feature points to identify the curved road edge points and the curved lane line points of the curved road point cloud segment.

[0102] In implementation, the road edge is the physical division of the road and the sidewalk, which appears as a distinct mutation edge in the terrain, for example, the height difference between the road edge stone and the road surface is large, forming a significant geometric curvature. Therefore, in the point cloud, the local curvature of the point cloud at the road edge is generally larger than that of the ordinary road surface or lane line. In the curved road processing, the actual road edge should be close to the virtual road edge position set by the vehicle pose expansion. If the centroid position of a point cloud is close to the left and right virtual road edge boundaries, it means that it is spatially located near the estimated edge of the road, rather than in the middle of the road. Therefore, only when the local curvature is large and the distance to the virtual road edge is close, the position distribution of the road edge is met, so that it is inferred that the point is a curved road edge point, rather than an ordinary ground noise point or a lane line point. By performing curvature calculation on the candidate feature points and combining feature parameters such as the distance from the centroid to the virtual road edge, the curved road edge points and the curved lane line points can be distinguished, and the accuracy of feature recognition can be improved. For example, a local neighborhood with a radius of 0.3 meters can be extracted for each candidate feature point, and the local curvature is calculated. If the local curvature is greater than 0.05 and the distance from the centroid to the left and right virtual road edges is less than a set range (such as 0.5 meters), it is determined to be a curved road edge point. If the local curvature is less than 0.05 and located in the middle area of the virtual road edge, it is determined to be a curved lane line point.

[0103] As an optional implementation, Figure 7 A flowchart of a curvature analysis method provided by an embodiment of the present application is shown in Figure 7 As shown in the flowchart of the curvature analysis method provided by an embodiment of the present application, the specific steps of identifying the curved road edge points and the curved lane line points of the curved road point cloud segment in step 603 are as follows:

[0104] In step 701, the candidate feature points are clustered.

[0105] In implementation, in actual point cloud data, candidate feature points are distributed in the form of locally dense clusters, corresponding to different physical structures (such as curbs, lane lines or other facilities). By clustering the candidate feature points, spatially connected or adjacent point clouds can be grouped together, which facilitates subsequent classification and identification based on overall features (such as centroid position, local curvature), thereby improving the accuracy and stability of the extraction of curved curbs and curved lane line points. For example, algorithms such as DBSCAN can be used to cluster candidate feature points. Set the neighborhood radius parameter to 0.3 meters and the minimum cluster point threshold to 10 points. Execute algorithms such as DBSCAN to divide point clouds with close distances and density requirements into different cluster blocks, and remove isolated points as noise points.

[0106] Step 702: For each cluster block, calculate the distance between the centroid position of the cluster block and the virtual curb and the point cloud curvature, and distinguish and summarize the point clouds corresponding to the distance and point cloud curvature into curved curb points and curved lane line points according to the preset classification rules.

[0107] In implementation, the spatial distribution positions and local curvature characteristics of different cluster blocks reflect the type of road elements to which they belong. Generally, the centroid of the curb point cloud cluster block is close to the virtual curb boundary and has a large local curvature, while the centroid of the lane line point cloud cluster block is located in the middle area of ​​the virtual curb and has a small curvature. By calculating the nearest distance from the centroid of each cluster block to the virtual curb and the mean curvature, and distinguishing them according to classification rules (such as the combination of curvature threshold and distance threshold), the curved curb points and curved lane line points can be effectively classified. For example: for each cluster block, calculate the spatial centroid position of all points, and find the nearest distance from the centroid to the left and right virtual curbs. Then, calculate the average value of the local curvature of each point in the cluster block. Classification rules can be set: if the average curvature of the cluster block is greater than 0.05 and the distance from the center of mass to the virtual curb on either side is less than 0.5 meters, it is judged as a curved curb point cluster block; if the average curvature is less than 0.05 and the center of mass position is close to the center area of ​​the virtual curb (such as within ±1 meter from the center), it is judged as a curved lane line point cluster block, thereby summarizing cluster blocks of different categories to generate curved curb point sets and curved lane line point sets.

[0108] Step 604 : Fit the curved road curb points and the curved lane line points into the curved road curb and the curved lane line, respectively.

[0109] In implementation, since the road edge and lane line of the curved road region are in curve form, curve fitting can be performed on the extracted curved road edge points and curved lane line points respectively. The fitting operation can smooth the local discrete point noise and form continuous and smooth road boundary lines and lane lines. For example, quadratic curve (such as parabola) or spline curve (such as cubic B-spline) can be used to fit the curved road edge points and curved lane line points respectively. In the fitting process, ceres g2o and other methods can be applied, and the RANSAC method can be used to exclude abnormal points to improve the stability and accuracy of the fitting results.

[0110] In step 105, a high-precision map of a vehicle driving region is generated according to the straight road edge, the curved road edge, the straight lane line and the curved lane line.

[0111] In implementation, the extracted different road features can be spliced into a complete continuous road network, and feature fitting, interval sampling and voxel filtering processing need to be performed to ensure that the map data has sufficient geometric accuracy and meets the requirements of sparse and regular point distribution, facilitating subsequent navigation, path planning and other applications. For example, linear or curve fitting can be performed on the straight road edge and the curved road edge to obtain continuous road edge lines, and the straight lane line and the curved lane line are fitted to obtain continuous lane lines. Then, the road edge and the lane line are discretely sampled at an interval of 0.5 meters, and a voxel filter (voxel size 0.2 meters) is used to sparsify the point cloud, and finally a standard format high-precision map data file is generated.

[0112] As an optional implementation, Figure 8 A flowchart of a high-precision map generation method provided by an embodiment of the present application is shown in Figure 8 As shown in the figure, the specific steps of generating a high-precision map of a vehicle driving region according to the straight road edge, the curved road edge, the straight lane line and the curved lane line in step 105 are as follows:

[0113] In step 801, the straight road edge and the curved road edge are spliced.

[0114] In implementation, since the straight road edge and the curved road edge are extracted in different road sections respectively, in order to ensure the continuity of the road boundary, they can be spliced in the order of spatial position to eliminate the boundary break between different sections, so that the overall road edge line presents continuous and smooth characteristics, meeting the requirements of high-precision map. For example, the straight road edge and the curved road edge can be arranged according to the vehicle driving order to determine the starting point and the ending point of each road edge section. By connecting the end point of the adjacent road edge section with the starting point of the next road edge section, a simple straight line connection or spline curve connection method is used to realize smooth transition, and seamless splicing of the road edge point cloud data is completed.

[0115] In step 802, the straight lane line and the curved lane line are spliced.

[0116] In practice, straight lane lines and curved lane lines correspond to different road morphologies, maintaining lane line continuity and a natural geometric transition during map generation. By splicing the two, a complete lane line network can be formed, ensuring continuous and accurate navigation guidance lines for path planning or autonomous driving applications. For example, the extracted straight lane line and curved lane line data can be sorted by vehicle pose timestamp, and the endpoints of adjacent lane line segments can be connected. During the splicing process, weighted linear interpolation or local curve smoothing methods can be used to eliminate breaks and sudden angle changes between lane line segments, thereby improving the continuity and smoothness of the overall lane line.

[0117] Step 803: Generate continuous road curbs and road lane lines using a preset fitting algorithm.

[0118] During implementation, there may still be local irregular distributions of the roadside points and lane line points after splicing. In order to generate continuous road elements that meet high-precision map standards, a fitting algorithm can be used to model the discrete point cloud as a whole, so that the road boundaries and lane lines present regular, continuous linear or curved shapes, thereby improving geometric accuracy and overall consistency. For example, a spline curve fitting algorithm (such as cubic B-spline fitting) can be applied to the spliced ​​road side points and lane line points respectively. During the fitting process, the control point spacing is set to 2 meters, the fitting error tolerance is not more than 5 centimeters, and the fitting results are expressed in the form of a continuous curve to facilitate subsequent loading and rendering by the map system.

[0119] In step 804 , the point clouds on the road curbs and road lane lines are subjected to interval sampling and voxel filtering to be converted into discrete scattered points with uniform spacing, so as to generate a high-precision map of the vehicle driving area.

[0120] In practice, the distribution density of the road boundary and lane line data points after continuous fitting is different, and direct use may lead to redundant or irregular point sets. By performing interval sampling on the fitting curve and combining it with voxel filter for unified sparse processing, the map data point cloud can have a regular spacing distribution, reducing the data volume while maintaining the road geometry accuracy, facilitating subsequent efficient storage, transmission and navigation. For example: on the fitted road curb and lane line curve model, equidistant sampling is performed at a fixed interval of 0.5 meters to generate a uniformly distributed discrete point set. The voxel filtering algorithm (voxel size 0.2 meters) is then used to thin out the sampled point set, remove over-dense points, and retain the center point, thereby forming regular, uniform, and high-precision map-compliant final road boundary and lane line data.

[0121] As an optional implementation, a plurality of local high-precision maps respectively generated by a plurality of vehicles in different vehicle driving areas can be acquired, and the high-precision maps of the plurality of vehicles are spliced and error-corrected by a preset adaptive splicing algorithm to obtain a global high-precision map of the plurality of vehicle driving areas.

[0122] In implementation, since the map generated by a single vehicle in a limited driving area is a local map, there is a problem of insufficient spatial coverage. By simultaneously mapping in different areas by multiple vehicles and transmitting data in real time through a cloud platform, the map data of multiple vehicles is used for collaborative splicing and optimization. In order to ensure the consistency and accuracy of the overall map, an adaptive splicing algorithm can be used to dynamically adjust the coordinate system registration parameters according to the overlapping area features between local maps, and to perform layer alignment and error correction. The adaptive splicing can effectively solve the problem of accumulated positioning error caused by different vehicle data sources, and generate a global high-precision map with wide coverage, good continuity and high accuracy. For example, local high-precision map data generated by a plurality of vehicles during driving can be collected, and preliminary registration can be performed according to the time stamp and geographical position information. An algorithm such as ICP can be used to match the point clouds of the local maps in the overlapping area, and to calculate the optimal translation and rotation matrix. According to the matching error size of the overlapping area, the coordinate system of the local map is adaptively adjusted to make the splicing boundary transition smoothly. Finally, a plurality of local maps are merged into a unified global map file, and the splicing gap is eliminated through post-processing to improve the consistency and accuracy of the overall map.

[0123] The embodiment of the application provides a high-precision mapping method, which comprises the following steps: collecting point cloud data by means of a laser radar during vehicle driving, and recording each frame of point cloud data and corresponding vehicle pose information. According to the vehicle pose information, the point cloud data is divided into straight line point cloud segments and curved road point cloud segments. Feature extraction is performed on the straight line point cloud segments to determine straight line road edges and straight line lane lines. Feature extraction is performed on the curved road point cloud segments to determine curved road edges and curved road lane lines. According to the straight line road edges, the curved road edges, the straight line lane lines and the curved road lane lines, a high-precision map of a vehicle driving area is generated. By collecting point cloud data in real time by means of a laser radar during vehicle driving, and synchronously recording the vehicle pose information corresponding to each frame of point cloud, the consistency of the point cloud data and the spatial motion state of the vehicle is ensured, which helps to improve the accuracy of subsequent spatial registration. In the point cloud data processing stage, according to the vehicle pose information, the point cloud data is automatically divided into straight line point cloud segments and curved road point cloud segments by accumulating the change of the rotation angle within a certain time or distance. By this segmentation method based on the change of the motion trajectory, the road types can be effectively distinguished, and the misjudgment problem caused by the traditional method based on the static point cloud density or simple curvature estimation is avoided, and the accuracy of road type identification is improved. For the straight line point cloud segments, the straight line road edge points are extracted by taking the vehicle center axis as the reference, combining the point cloud height jump characteristics and the horizontal direction angle judgment, and further screening the lane line points between the road edges based on the reflectivity threshold, so that the road boundaries and lane lines of the straight line road segments are accurately extracted. For the curved road point cloud segments, the virtual road edges are constructed on the left and right sides of the vehicle, and the road edge points and lane line points in the curved road area are comprehensively screened and clustered by combining the point cloud reflectivity, density and curvature characteristics. By introducing the curvature analysis, the curved road edges can be more accurately identified, and different types of curved road lane lines can be effectively distinguished, and the reliability and accuracy of feature extraction in complex curved road conditions are improved. In the map generation stage, the extracted straight line road edges, curved road edges, straight line lane lines and curved road lane lines are respectively spliced and processed, and a fitting algorithm is used to generate continuous road edge lines and lane lines, and further through interval sampling and voxel filtering processing, the continuous line features are converted into regularly distributed discrete point clouds, so that the generated high-precision map has high geometric precision, good data sparseness and processing efficiency. Therefore, by dynamically dividing the road types according to the vehicle pose information, extracting the road edge and lane line features based on the geometric characteristics and reflectivity characteristics, and generating the continuous road network through efficient data processing and fitting, the problems of difficult road feature extraction, large recognition error of straight line and curved road areas, discontinuous map and low overall generation efficiency in the background art are effectively solved.

[0124] It should be understood that, although Figures 1 to 8The steps in the flowcharts are shown in sequence according to the arrows, but the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the steps are not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, Figures 1 to 8 At least a part of the steps in the flowcharts can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least a part of other steps or steps or stages in other steps.

[0125] It can be understood that the same / similar parts between each embodiment of the above method in the specification can be mutually referred to, and each embodiment focuses on the differences from other embodiments, and the related parts can be referred to the description of other method embodiments.

[0126] In one embodiment, a computer device is provided, as shown in Figure 9 The computer device includes a memory and a processor, and the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the above high-precision mapping method.

[0127] In one embodiment, a computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above high-precision mapping method.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0129] It should be noted that, in this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or equipment including the element.

[0130] It should also be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in this application are information and data authorized by the user or authorized by all parties.

[0131] The various embodiments in the specification are described in a related manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, the system embodiments are described simply because they are basically similar to the method embodiments, and the related parts can be referred to the part of the method embodiments.

[0132] The technical features of the above embodiments can be combined in any manner. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0133] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. A high-precision map drawing method, characterized in that: The method comprises: During the vehicle's driving process, point cloud data is collected through the laser radar, and each frame of the point cloud data and the corresponding vehicle posture information are recorded; Dividing the point cloud data into straight line point cloud segments and curve point cloud segments according to the vehicle posture information; Extracting features from the straight point cloud segments to determine straight curbs and straight lane lines; Extracting features from the curve point cloud segment to determine the curve curb and the curve lane line; A high-precision map of the vehicle driving area is generated based on the straight roadside, the curved roadside, the straight lane lines and the curved lane lines.

2. The method according to claim 1, characterized in that The laser radar is a mechanical rotating laser radar or a solid-state dot matrix laser radar.

3. The method according to claim 1, characterized in that The step of dividing the point cloud data into straight point cloud segments and curved point cloud segments according to the vehicle posture information includes: Accumulating the vehicle posture information and calculating the cumulative rotation angle within a preset time or distance; When the accumulated rotation angle is lower than a preset maximum allowable straight line angle threshold, determining the point cloud data corresponding to the vehicle posture information as the straight line point cloud segment; When the accumulated rotation angle is greater than a preset curve angle threshold, the point cloud data corresponding to the vehicle posture information is determined as the curve point cloud segment.

4. The method according to claim 1, wherein The method for determining a straight road edge comprises: Dividing the point cloud data of the straight point cloud segment into left and right sides based on the central axis of the vehicle; For the point cloud data on each side, the straight roadside point is identified based on the height jump and the preset angle threshold; The straight roadside points are fitted into the straight roadside.

5. The method according to claim 4, characterized in that The step of identifying the straight roadside point based on the point cloud data on each side and the height jump and the preset angle threshold comprises: For each point in the point cloud data, calculate the angle between the height difference between the point and the nth point on its left and right and the origin of the lidar; If the angle is greater than the angle threshold, the point is determined to be a horizontal feature point; For each of the horizontal feature points, combined with the height variance, horizontal feature points with height jumps are screened out as straight roadside points.

6. The method according to claim 1, characterized in that The method for determining a straight lane line includes: Extracting point cloud data of the area between the straight road edges; In the point cloud data of the area between the straight roadside edges, point clouds having a reflectivity greater than a preset first reflectivity threshold are determined as straight lane line points; The straight lane line points are fitted into the straight lane line.

7. The method according to claim 1, characterized in that The feature extraction of the curve point cloud segment to determine the curve curb and the curve lane line includes: Expanding a virtual curb within the curve point cloud segment based on the vehicle posture information and a preset curb width; Determining point clouds located between the virtual curbs and having a reflectivity greater than a preset second reflectivity threshold and a point cloud density greater than a preset point cloud density threshold as candidate feature points; Performing curvature analysis on the point cloud of the candidate feature points to identify the curve curb points and curve lane line points of the curve point cloud segment; The curved road curb point and the curved lane line point are fitted as the curved road curb and the curved lane line respectively.

8. The method according to claim 7, characterized in that The performing curvature analysis on the point cloud of the candidate feature points to identify the curved road curb points and curved lane line points of the curved point cloud segment includes: Clustering the candidate feature points; For each cluster block, the distance between the centroid position of the cluster block and the virtual curb and the point cloud curvature are calculated, and the point clouds corresponding to the distance and point cloud curvature are distinguished and summarized as the curved curb points and the curved lane line points according to the preset classification rules.

9. The method according to claim 1, characterized in that Generating a high-precision map of the vehicle driving area according to the straight roadside, the curved roadside, the straight lane line, and the curved lane line includes: splicing the straight curb and the curved curb; splicing the straight lane line and the curved lane line; Generate continuous road curbs and road lane lines through a preset fitting algorithm; The point clouds on the road curb and the road lane lines are subjected to interval sampling and voxel filtering processing to be converted into discrete scattered points with consistent spacing, so as to generate a high-precision map of the vehicle driving area.

10. The method according to claim 1, characterized in that The method further comprises: Obtaining local high-precision maps generated by multiple vehicles in different vehicle driving areas; The high-precision maps of multiple vehicles are layered and error corrected through a preset adaptive stitching algorithm to obtain a global high-precision map of the driving area of ​​multiple vehicles.

Images