A laser radar-based passable area detection method

Abstract

The application provides a laser radar-based passable area detection method, characterized in that the method comprises the following steps: acquiring single-frame point cloud data of a laser radar; extracting a straight line feature in the single-frame point cloud data, and acquiring road boundary point cloud data and road boundary straight line fitting information according to the straight line feature; segmenting the single-frame point cloud data according to the road boundary point cloud data and the road boundary straight line fitting information to obtain a first passable area; extracting obstacle point cloud data according to adjacent point angle information, and obtaining a second passable area according to the first passable area and the obstacle point cloud data. The scheme provided by the application can realize stable detection of a passable area in a structured scene and a semi-structured scene.

Description

Technical Field

[0001] This application relates to the field of detection, and more particularly to a method for detecting passable areas based on lidar. Background Technology

[0002] In recent years, with the gradual development of autonomous driving technology, robotaxis have moved from testing and research to trial operation. Accessible area detection is a key technology in the field of autonomous driving and has important applications in robotaxis. Currently, the more mature algorithms and applications are mainly designed for flat roads with clear signs and markings. However, in the actual operating environment of robotaxis, it is inevitable that they will enter scenarios with poor lighting, unclear signs and markings, ambiguous road boundaries, or certain slopes and bumps. Methods capable of robustly detecting accessible areas in both structured and semi-structured scenarios are crucial for improving the reliability of robotaxis operation.

[0003] Mainstream sensors used for feasible region detection include LiDAR and cameras. Camera-based feasible region detection methods mainly extract feasible regions by extracting features such as lane lines, road edge lines, and road surface texture. These methods require the scene to have good feature information, and due to the limitations of sensor characteristics, they are greatly affected by changes in lighting and weather conditions, and have poor adaptability to different environmental scenarios.

[0004] In recent years, LiDAR-based autonomous vehicle accessibility detection technology has developed rapidly, and can be divided into two technical directions. One is rule-based detection methods, such as grid-based methods and methods based on point location distribution features. The second is deep learning-based methods. In the first type of methods, road surface extraction mainly relies on height information, which is not very adaptable to undulating road environments; moreover, these methods do not adequately consider the temporal correlation between multiple frames of data. In contrast, deep learning-based methods have high computational resource requirements and require a large amount of manpower for labeling training samples.

[0005] Previous research on feasible region detection algorithms has mainly focused on two types of scenarios: structured road scenarios, where feasible region detection primarily focuses on road boundary extraction, road surface segmentation, and the detection of dynamic and static obstacles; and unstructured scenarios, where feasible region detection primarily focuses on the detection and recognition of positive and negative obstacles. The road surface undulation and bumpiness in semi-structured environments are less pronounced than in unstructured environments such as off-road scenarios, therefore, the technical approach is more similar to feasible region detection in structured scenarios.

[0006] Rule-based drivable area detection, such as the patent "Drivable Area Road Surface Extraction Method and Device" (202011139074.5), discloses a method and device for extracting drivable area road surfaces. The main steps include acquiring road surface point clouds in a vehicle coordinate system using LiDAR and acquiring road surface edge points in the same system using a high-precision map; dividing the road surface into N rectangles based on the edge points; acquiring point clouds whose height is greater than a preset height threshold and are located within the N rectangles; fitting the road surface based on the point clouds to obtain the roll angle and pitch angle of the fitted road surface; and using the fitted road surface's roll angle less than a first preset angle and its pitch angle less than a second preset angle as the current road surface. This patent addresses the algorithm's robustness by incorporating high-precision maps to provide road surface edge points for segmentation. Therefore, the road segmentation result is highly dependent on map accuracy, and it does not consider the correlation between multiple frames of point cloud data when analyzing road drivability.

[0007] The patent "A Method and System for Passable Area Detection in Complex Environments" (202011164865.3) discloses a method for detecting passable areas in complex environments, with the following main steps: 1. Constructing a point cloud deep learning detection subsystem for passable areas, including a road type recognition network and a passable area detection network; 2. Constructing a traditional method detection subsystem for passable areas, including a road type recognition module and a passable area detection module; 3. The point cloud deep learning detection subsystem and the traditional method detection subsystem are directly connected through the underlying interface of the deep learning framework; 4. Newly acquired point cloud data from the vehicle-mounted LiDAR is preferentially input into the point cloud deep learning detection subsystem. This method improves the adaptability of the detection algorithm to different environments by combining deep learning algorithms with traditional detection algorithms. However, because the fusion method of the two methods in this patent is relatively simple, the complementary advantages brought by the fusion are not obvious. Furthermore, it does not alleviate the dependence of deep learning detection algorithms on samples, nor does it improve the adaptability of traditional detection algorithms to different environments. Summary of the Invention

[0008] In view of this, this application provides a passable area detection method based on lidar, which can stably detect passable areas in both structured and semi-structured scenes.

[0009] Specifically, this application is implemented through the following technical solution:

[0010] According to a first aspect of this application, a method for detecting passable areas based on lidar is provided, the method comprising:

[0011] Acquire single-frame point cloud data from LiDAR;

[0012] Extract the straight line features from the single frame point cloud data, and obtain the road boundary point cloud data and road boundary straight line fitting information based on the straight line features;

[0013] Based on the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data is segmented to obtain the first passable area.

[0014] Obstacle point cloud data is extracted based on the angle information of adjacent points, and a second passable area is obtained based on the first passable area and the obstacle point cloud data.

[0015] Optionally, the step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes,

[0016] The disordered point cloud data is organized into ordered point cloud data based on the radar's vertical and horizontal angular resolutions. The single-frame point cloud data of the lidar includes the three-dimensional coordinate information of all returned laser points.

[0017] Optionally, the step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes,

[0018] Based on the fuzzy line segment method, the straight line features in the single frame point cloud data are extracted line by line, and the straight line features are filtered and clustered based on the line segment length, slope, intercept and width information.

[0019] Optionally, the step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes,

[0020] Extract data from a specific line sequence in ordered point cloud data, project it onto a plane, and calculate the squared difference between the position of the starting point and the n points before and after it. Select the points whose position variance is greater than the first threshold as the segmentation points, and divide the single-line sequence point cloud data into several discrete curve point clouds. For each discrete curve point cloud, fit it based on the fuzzy line segment method, and record the fuzzy line segment fitting data of the discrete curve segment as the feature value.

[0021] Optionally, the fitting of each discrete curve point cloud segment based on the fuzzy line segment method includes,

[0022] If the discrete point set (x, y) of the discrete curve point cloud satisfies the following for a preset discrete line L(a, b, u),

[0023] If u ≤ ax - by < u + w, and the length of the discrete curve is greater than the second threshold, then the point cloud of the discrete curve segment is added to the candidate point cloud of the road boundary line, and the fuzzy line segment fitting data of the discrete curve segment is recorded as a feature value. The feature value includes the start and end points, slope, intercept, and width, where the width is the maximum straight-line distance between the candidate point cloud and the fitted line. Wherein, a and b are the slope parameters of the discrete line L, u is the intercept of the discrete line L, and w is the third threshold.

[0024] Optionally, the data of all line sequences are fitted using the fuzzy line segment method to segment and obtain all point clouds with straight line features in a single frame of point cloud data, as well as the feature values ​​of their fitted line segments. Thresholds are set for the slope and intercept of the line segments. If the difference in the slope and intercept of the straight line features contained in the point clouds of different line sequences is within the set threshold range, the straight line features are merged and the start and end point information is updated. The merged straight line features are analyzed, and line segments with approximately equal slopes and differences from the vehicle's driving direction that meet the threshold requirements are selected as lane boundary lines. The discrete curve point cloud corresponding to the lane boundary line is extracted as the lane boundary point cloud.

[0025] Optionally, based on the road boundary point cloud data and the road boundary line fitting information, the single frame point cloud data is segmented to obtain the first passable area, including:

[0026] Based on the extracted lane boundary point cloud and the start and end points of the lane boundary line, the single frame point cloud data is segmented, the point cloud data outside the lane boundary line is segmented, and the remaining point cloud is taken as the feasible area point cloud to obtain the first passable area.

[0027] Optionally, before extracting obstacle point cloud data based on the angle information of adjacent points and obtaining the second passable area based on the first passable area and the obstacle point cloud data, the method further includes...

[0028] The inertial measurement unit data and lidar data are synchronized based on GPS timing.

[0029] Based on inertial measurement unit data, the point cloud data of the current frame and the previous frame are registered; and the coordinate transformation relationship is recorded. The line segment features in the point cloud data of the previous frame are transformed and compared with the line segment features of the current frame. If the slope difference is greater than the fourth threshold, the superposition of the two frames of data is abandoned and a single frame of data is used for obstacle extraction. If the slope difference is less than the fourth threshold, the two frames of data are superimposed for obstacle detection.

[0030] Optionally, obstacle point cloud data is extracted based on the angle information of adjacent points. Based on the first passable area and the obstacle point cloud data, the second passable area is obtained, including...

[0031] The single-frame point cloud data is divided into several rays according to its horizontal angle;

[0032] Select a point cloud on a certain ray, and calculate the angular relationship between three consecutive points in sequence to determine the feasibility. The specific feasibility judgment criterion is defined as ΔD1=E(α0)-(α1+θ).

[0033] ΔD2=α1+θ-α0; In the formula, E(α0) represents the expected value of the laser beam hitting the ground, and α0 can be calculated from the laser installation height and the vertical angle of the line sequence of the point.

[0034] If ΔD1≈0 and ΔD2≈0, then it is a flat and passable area. If ΔD1≈0, and

[0035] If ΔD2>0, then the point closest to the vehicle body among the three points will protrude; if ΔD1>0 and ΔD2<0, then the middle point among the three points will affect the protrusion of the vehicle body; if ΔD1<0 and ΔD2>0, then the farthest point among the three points will affect the protrusion of the vehicle body; where θ is the vertical angular resolution of the lidar, and α0 and α1 are the angles between the laser beam and the ground.

[0036] Optionally, the point cloud on each ray is screened according to the feasibility judgment criteria to obtain the obstacle point cloud. The obstacle point cloud that affects vehicle driving is segmented from the first passable area to obtain the second passable area.

[0037] As described above, by extracting straight line features from single-frame point cloud data, road boundary point cloud data and road boundary straight line fitting information are obtained based on the straight line features; the single-frame point cloud data is segmented based on the road boundary point cloud data and road boundary straight line fitting information to obtain a first passable area; obstacle point cloud data is extracted based on the angle information of adjacent points, and a second passable area is obtained based on the first passable area and obstacle point cloud data, thereby obtaining the vehicle passable area, achieving stable detection of feasible areas in both structured and semi-structured scenes. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating an exemplary embodiment of a passable area detection method based on lidar.

[0039] Figure 2 This is a schematic diagram illustrating another passable area detection method based on lidar, as shown in an exemplary embodiment of this application.

[0040] Figure 3 This is a schematic diagram illustrating the angle information of adjacent points in an exemplary embodiment of this application.

[0041] Explanation of reference numerals in the attached figures:

[0042] 301 - Vehicles; 302 - LiDAR. Detailed Implementation

[0043] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0044] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0045] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0046] See Figure 1 , Figure 1 This is a schematic diagram of a passable area detection method based on lidar, which includes the following steps.

[0047] Step 101: Obtain single-frame point cloud data from the lidar.

[0048] This single-frame point cloud data can be acquired by installing a lidar on the top of the vehicle. The single-frame point cloud data includes the three-dimensional coordinate information of all returned lidar points.

[0049] Step 102: Extract the straight line features from the single frame point cloud data, and obtain the road boundary point cloud data and road boundary straight line fitting information based on the straight line features.

[0050] Based on the fuzzy line segment method, straight line features are extracted from the single-frame point cloud line by line. Based on information such as line segment length, slope, and intercept, the straight line features are filtered and clustered to obtain road boundary point cloud data and road boundary straight line fitting information.

[0051] Step 103: Based on the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data is segmented to obtain the first passable area.

[0052] Based on the straight line features obtained in step 102, after obtaining the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data of the lidar is segmented according to the road boundary point cloud data and the road boundary straight line fitting information. The point cloud data outside the road boundary line is segmented, and the remaining point cloud is taken as the feasible area point cloud to obtain the first passable area. The first passable area is the area within the road boundary line, which still includes the obstacle area within the road.

[0053] Step 104: Extract obstacle point cloud data based on the angle information of adjacent points, and obtain the second passable area based on the first passable area and the obstacle point cloud data.

[0054] After obtaining the obstacle point cloud data, the obstacle point cloud is segmented from the first passable area to obtain the second passable area. The second passable area has already separated the obstacle area from the first passable area, thus completing the vehicle passable area detection.

[0055] This embodiment extracts road boundaries from the original point cloud data using the fuzzy line segment method to segment the point cloud data. Compared with the ground point cloud segmentation method based on height threshold, it is more applicable to sloping roads. Furthermore, it extracts obstacle point cloud data based on the angle information of adjacent points, which further improves the accuracy of detection.

[0056] See Figure 2 , Figure 2 This is a schematic diagram of another passable area detection method based on lidar, which includes the following steps.

[0057] Step 201: Obtain single-frame point cloud data from the lidar.

[0058] Step 202: Extract the straight line features from the single frame point cloud data, and obtain the road boundary point cloud data and road boundary straight line fitting information based on the straight line features.

[0059] Step 203: Based on the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data is segmented to obtain the first passable area.

[0060] Based on the straight line features obtained in step 202, after obtaining the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data of the lidar is segmented according to the road boundary point cloud data and the road boundary straight line fitting information. The point cloud data outside the road boundary line is segmented, and the remaining point cloud is taken as the feasible area point cloud to obtain the first passable area. The first passable area is the area within the road boundary line, which still includes the obstacle area within the road.

[0061] Step 204: Based on the inertial measurement unit data, register two consecutive point cloud frames; based on the registration result, select whether to use the two point cloud frames for obstacle detection.

[0062] Based on inertial measurement unit (IMU) data and a vehicle kinematics model, two consecutive frames of point cloud data are registered. The registration result determines whether to use both frames for obstacle detection. The IMU can be mounted on a vehicle, and the vehicle kinematics model is a commonly used model.

[0063] Step 205: Extract obstacle point cloud data based on the angle information of adjacent points, and obtain the second passable area based on the first passable area and the obstacle point cloud data.

[0064] Step 202 further includes:

[0065] Step 202-1 involves organizing the unordered single-frame point cloud data from the lidar into ordered point cloud data based on the lidar's vertical and horizontal angular resolution. The single-frame point cloud data from the lidar includes the three-dimensional coordinate information of all returned lidar points.

[0066] Step 202-2: Extract data of a specific line sequence from the ordered point cloud, project it onto the XY plane, and starting from the initial scan point, calculate the squared differences between the position of this point and the n points before and after it. The calculation formula is as follows: Points with large positional variance are selected as segmentation points to divide the single-line sequence point cloud data into several discrete curve point clouds. For each discrete curve point cloud segment, a fuzzy line segment method is used for fitting. If the definition of a fuzzy line segment is satisfied, that is, if the discrete point set (x, y) of the discrete curve point cloud satisfies the preset discrete line L(a, b, u),

[0067] If u ≤ ax - by < u + w, and the length of the discrete curve is greater than a specified threshold, then the point cloud of this discrete curve segment is added to the candidate point cloud of the road boundary line, and the fuzzy line segment fitting data of this discrete curve segment is recorded as a feature value. This feature value includes the start and end points, slope, intercept, and width, where the width is the maximum straight-line distance between the candidate point cloud and the fitted line. Here, a and b are the slope parameters of the discrete line L, determining the slope of the discrete line L; u is the intercept of the discrete line L; and w is the threshold that needs to be satisfied.

[0068] Step 202-3: Perform the operation of step 202-2 on all line sequences to segment and obtain all point clouds with straight line features in the single frame point cloud data, as well as the feature values ​​of their fitted line segments. Set thresholds for the slope and intercept of the line segments. If the difference in the slope and intercept of the straight line features contained in the point clouds of different line sequences is within the set threshold range, then merge the straight line features and update the start and end point information. Analyze the merged straight line features and select line segments with approximately equal slopes and whose difference from the vehicle's driving direction meets the threshold requirement as lane boundary lines. Extract the discrete curve point clouds contained in them as lane boundary point clouds.

[0069] Step 204 further includes:

[0070] Step 204-1: Synchronize the inertial measurement unit data and lidar data based on GPS timing.

[0071] Step 204-2: Based on the inertial measurement unit data and the vehicle kinematics model, register the point cloud data of the current frame with the previous frame; and record the coordinate transformation relationship. Perform coordinate transformation on the straight line segment features in the previous frame point cloud data and compare them with the straight line segment features of the current frame. If the slope difference is greater than a specified threshold, discard the superposition of the two frames of data and use the single frame data for obstacle extraction. If the slope difference is less than a specified threshold, superimpose the two frames of data for obstacle detection.

[0072] Step 205 further includes:

[0073] Step 205-1: Divide the single-frame point cloud data into several rays according to its horizontal angle.

[0074] Step 205-2: Select a point cloud on a certain ray, and calculate the angle relationship between three consecutive points to determine feasibility. The specific feasibility judgment criteria are defined as follows: ΔD1 = E(α0) - (α1 + θ), ΔD2 = α1 + θ - α0; In the formula, E(α0) represents the expected value when the laser beam hits the ground, and α0 can be calculated from the laser installation height and the vertical angle of the line sequence of the points. If ΔD1 ≈ 0 and ΔD2 ≈ 0, it is a flat and passable area; if ΔD1 ≈ 0 and ΔD2 > 0, the point closest to the vehicle body is protruding; if ΔD1 > 0 and ΔD2 < 0, the middle point among the three points affects the vehicle body protrusion; if ΔD1 < 0 and ΔD2 > 0, the farthest point among the three points affects the vehicle body protrusion; where θ is the vertical angle resolution of the lidar, and α0 and α1 are the angles between the laser beam and the ground. See details... Figure 3 Using this feasibility assessment criterion, point clouds along the ray that affect vehicle movement were selected.

[0075] Step 205-3: Based on the feasibility judgment criteria in step 205-2, the point cloud on each ray is screened to obtain the obstacle point cloud. The obstacle point cloud that affects vehicle driving is segmented from the first passable area to obtain the second passable area.

[0076] This embodiment extracts road boundaries from the original point cloud data using the fuzzy line segment method to segment the point cloud data. Compared with ground point cloud segmentation methods based on height thresholds, it is more applicable to sloping roads. Obstacle point cloud data is extracted based on the angle information of adjacent points, further improving the detection accuracy. By extracting obstacle point clouds from multiple frames of point cloud data, the search range for judging obstacle point clouds based on the angle between continuous points is reduced, thus lowering the computational load. Furthermore, by extracting obstacle point clouds from multiple frames of point cloud data, the impact of differences in data density from point clouds collected at a single location on the extraction results is reduced, improving the robustness of the extraction results.

[0077] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting passable areas based on lidar, characterized in that, The method includes: Acquire single-frame point cloud data from LiDAR; Extract the straight line features from the single frame point cloud data, and obtain the road boundary point cloud data and road boundary straight line fitting information based on the straight line features; Based on the road boundary point cloud data and the road boundary straight line fitting information, the single frame point cloud data is segmented to obtain the first passable area. Obstacle point cloud data is extracted based on the angle information of adjacent points, and a second passable area is obtained based on the first passable area and the obstacle point cloud data; Before extracting obstacle point cloud data based on the angle information of adjacent points and obtaining the second passable area based on the first passable area and obstacle point cloud data, the process includes synchronizing the inertial measurement unit data and lidar data based on GPS timing; registering the point cloud data of the current frame with the previous frame based on the inertial measurement unit data; recording the coordinate transformation relationship; transforming the coordinates of the straight line segment features in the previous frame point cloud data and comparing them with the straight line segment features in the current frame; if the slope difference is greater than the fourth threshold, the superposition of the two frames of data is abandoned, and a single frame of data is used for obstacle extraction; if the slope difference is less than the fourth threshold, the two frames of data are superimposed for obstacle detection.

2. The method according to claim 1, characterized in that, The step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes organizing the disordered point cloud data into ordered point cloud data based on the radar's vertical and horizontal angular resolution. The single-frame point cloud data of the lidar includes the three-dimensional coordinate information of all returned laser points.

3. The method according to claim 1, characterized in that, The step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes: extracting straight line features from the single-frame point cloud data line by line based on the fuzzy line segment method, and filtering and clustering the straight line features based on the line segment length, slope, intercept and width information.

4. The method according to claim 2, characterized in that, The step of extracting straight line features from the single-frame point cloud data and obtaining road boundary point cloud data and road boundary straight line fitting information based on the straight line features includes: extracting data of a certain line sequence from the ordered point cloud data, projecting it onto a plane, calculating the squared difference of the position between the starting scanning point and the n points before and after it in sequence, selecting points with a position variance greater than a first threshold as segmentation points, dividing the single-line sequence point cloud data into several segments of discrete curve point cloud, fitting each segment of discrete curve point cloud based on the fuzzy line segment method, and recording the fuzzy line segment fitting data of the discrete curve segment as feature values.

5. The method according to claim 4, characterized in that, The fitting of each discrete curve point cloud segment based on the fuzzy line segment method includes: if the discrete point set (x, y) of the discrete curve point cloud satisfies u≤ax-by<u+w for a preset discrete line L(a, b, u), and the length of the discrete curve is greater than a second threshold, then the discrete curve point cloud segment is added to the candidate point cloud of the road boundary line, and the fuzzy line segment fitting data of the discrete curve segment is recorded as a feature value; the feature value includes the start and end points, slope, intercept, and width, where the width is the maximum straight-line distance between the candidate point cloud and the fitted line; where a and b are the slope parameters of the discrete line L, u is the intercept of the discrete line L, and w is the third threshold.

6. The method according to any one of claims 1 to 5, characterized in that, The data of all line sequences are fitted using the fuzzy line segment method to obtain all point clouds with straight line features in a single frame of point cloud data, as well as the feature values ​​of their fitted line segments. Thresholds are set for the slope and intercept of the line segments. If the difference in the slope and intercept of the straight line features contained in the point clouds of different line sequences is within the set threshold range, the straight line features are merged and the start and end point information is updated. The merged straight line features are analyzed, and line segments with approximately equal slopes and differences from the vehicle's driving direction that meet the threshold requirements are selected as lane boundary lines. The discrete curve point cloud corresponding to the lane boundary line is extracted as the lane boundary point cloud.

7. The method according to claim 6, characterized in that, Based on the road boundary point cloud data and the road boundary line fitting information, the single frame point cloud data is segmented to obtain a first passable area. This includes segmenting the single frame point cloud data based on the extracted lane boundary point cloud and the start and end points of the lane boundary lines, segmenting the point cloud data outside the lane boundary lines, and using the remaining point cloud as the feasible area point cloud to obtain the first passable area.

8. The method according to claim 1, characterized in that, Obstacle point cloud data is extracted based on the angle information of adjacent points. Based on the first passable area and the obstacle point cloud data, the second passable area is obtained. This includes dividing a single frame of point cloud data into several rays according to its horizontal angle; selecting point clouds on a certain ray, and sequentially calculating the angle relationship between three consecutive points to determine feasibility. The specific feasibility judgment criteria are defined as follows: ΔD1 = E(α0) - (α1 + θ), ΔD2 = α1 + θ - α0; where E(α0) represents the expected value when the laser beam hits the ground. The value α0 can be calculated from the laser installation height and the vertical angle of the line sequence of the points. If ΔD1≈0 and ΔD2≈0, it is a flat and passable area. If ΔD1≈0 and ΔD2>0, the point closest to the vehicle body among the three points is a protrusion. If ΔD1>0 and ΔD2<0, the middle point among the three points has a protrusion that affects the vehicle body. If ΔD1<0 and ΔD2>0, the farthest point among the three points has a protrusion that affects the vehicle body. Where θ is the vertical angle resolution of the laser radar, and α0 and α1 are the angles between the laser beam and the ground.

9. The method according to claim 8, characterized in that, Based on the feasibility judgment criteria, the point cloud on each ray is screened to obtain the obstacle point cloud. The obstacle point cloud that affects vehicle driving is segmented from the first passable area to obtain the second passable area.

Images