A roadside detection method based on multi-line laser radar

Abstract

The present invention provides a curb detection method based on a multi-line laser radar, comprising the following steps: S1, obtaining raw point cloud data; scanning the vehicle's surroundings using a three-dimensional laser radar to obtain raw point cloud data coordinates; S2, preprocessing the point cloud data; clustering and segmenting the acquired point cloud data to extract ground point cloud data, selecting a region of interest, and setting a corresponding threshold; S3, selecting candidate curb points; calculating curb feature information from the point cloud data within the region of interest, comparing it with a set threshold, and selecting candidate curb points; S4, fitting the curb points; performing a RANSAC algorithm and a least squares fitting method on the selected candidate curb points to obtain a curb model. The present invention can accurately extract curb information from the environment, significantly reducing the impact of environmental factors such as light and weather changes on the detection effect, has high robustness, low computational complexity, and can extract curb information from the environment in real time.

Description

Technical Field

[0001] The present invention belongs to the field of environmental perception technology, and in particular relates to a roadside detection method based on a multi-line laser radar. Background Art

[0002] Curb detection is a relatively stable road feature in urban environments. It provides critical road boundary information for outdoor navigation of autonomous vehicles (AGVs). It plays a role in vehicle positioning, path planning, and decision-making. It can also provide temporary positioning information for vehicles when GPS is unreliable. Depending on the sensor used, the curb detection problem can be divided into two methods: vision-based and lidar-based.

[0003] In "Lane Line Detection Algorithm Based on Quadratic Threshold Segmentation and Lane Width Matching", Hu et al. proposed a roadside detection method based on quadratic threshold segmentation of the Otsu method. However, the vision-based roadside detection algorithm is easily affected by factors such as light and weather. Three-dimensional lidar is a commonly used sensor for sensing the surrounding environment of AGVs. Although the features it obtains are not as dense as those of cameras, due to the characteristic of lidar actively emitting lasers, it is more robust to changes in lighting in the environment and has higher ranging accuracy. In "A Real-time Road Edge Extraction Algorithm Based on 3D Lidar", Li et al. proposed a real-time road edge extraction algorithm based on 3D lidar. However, this method does not consider buildings on both sides of the road and obstacles in the road, and has poor anti-interference ability. In "Research on Road Boundary and Obstacle Detection Based on Lidar", a four-line lidar is used as a sensor, and the interval collinear point features of the multi-line lidar are used to obtain roadside points. The obtained roadside points are clustered using the improved DPCA algorithm. However, the four-line lidar has fewer scanning lines and obtains limited environmental features. In addition, this method does not fully consider the spatial characteristics of the roadside, resulting in low detection accuracy and poor robustness.

[0004] At present, the roadside detection method based on vision or lidar has high requirements on the environment, poor robustness, high computational complexity, and cannot meet the real-time requirements; therefore, this patent application designs a roadside detection method based on multi-line lidar. Summary of the Invention

[0005] In view of this, the present invention aims to propose a roadside detection method based on multi-line lidar to solve the problems that the existing method is easily affected by weather, has poor robustness, high computational complexity, and is difficult to meet real-time requirements.

[0006] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0007] A roadside detection method based on a multi-line laser radar comprises the following steps:

[0008] S1. Obtaining original point cloud data: Scanning the vehicle's surroundings with a 3D laser radar to obtain the coordinates of the original point cloud data;

[0009] S2. Point cloud data preprocessing: cluster and segment the acquired point cloud data, extract ground point cloud data, select the region of interest, and set the corresponding threshold;

[0010] S3. Selecting candidate roadside points: Calculating roadside feature information from the point cloud data within the region of interest, comparing it with a set threshold, and selecting candidate roadside points;

[0011] S4. Fitting of roadside points: performing RANSAC algorithm fitting and least squares fitting on the selected roadside candidate points to obtain a roadside model.

[0012] Furthermore, step S2 specifically includes the following steps:

[0013] S21, point cloud projection; establishing a three-dimensional coordinate system with the three-dimensional laser radar as the origin, converting the original point cloud data represented by polar coordinates into point cloud data in the three-dimensional coordinate system, and projecting the point cloud data into an image containing distances. The image includes row and column values ​​of the point cloud and distance information, where the row value indicates the ring number where the laser point is located, and the column value indicates the number of points on the ring;

[0014] S22, extracting the ground; clustering and segmenting the point cloud data, where clustering treats points with similar features as points of the same type, and segmentation divides a frame of point cloud into ground points and non-ground points;

[0015] S23. Select an area of ​​interest; in the ground point cloud data, select an area of ​​interest and set a corresponding threshold.

[0016] Furthermore, step S21 specifically includes the following method:

[0017] A three-dimensional coordinate system is established with the center of the three-dimensional laser radar as the origin, the front of the vehicle as the positive direction of the x-axis, the left of the vehicle as the positive direction of the y-axis, and the top of the vehicle as the positive direction of the z-axis. The point cloud represented by the polar coordinate system is converted into a point cloud in the three-dimensional coordinate system according to formula (1):

[0018]

[0019] Where R is the distance information of point i, γ i is the angle in the vertical direction, θ i is the angle in the horizontal direction, x i 、y i 、z i is the coordinate value of point i converted to the Cartesian coordinate system;

[0020] Calculate the number of rings that each point in the point cloud data is in, and the sequence value of each point on the ring where it is located, as the row value row and column value column of the index point, respectively calculated by formula (2):

[0021]

[0022] in, γ i represents the scanning angle of point i in the vertical direction, x i ,y i , z i is the coordinate value of point i; γ b is the lowest vertical scanning angle of the sensor, γ res is the angular resolution of the sensor in the vertical direction; θ i Indicates the angle of the point in the horizontal direction, θ res It is the horizontal angle resolution, H represents the number of points on each laser beam, and the distance information of the points is obtained by calculate.

[0023] Furthermore, the specific method of step S22 is: clustering the point cloud using the region growing method, specifying a point as the search starting point, calculating the angle between the points in the x direction or y direction in its neighborhood and the point, and setting a threshold θ Threshold =60°, if the angle is greater than this value, it is considered that the neighboring point and the point are of the same type, and the search continues with the neighboring point as the center point until there is no point that meets the conditions; according to the feature that the angle between the points corresponding to the two adjacent rings on the ground scanned by the laser radar is small in the vertical direction, the feature value is used. Segment ground points, where Δx, Δy, and Δz are the differences in the x, y, and z coordinates of corresponding points on adjacent rings. When the angle α is less than 15°, the point is considered a ground point.

[0024] Furthermore, in step S23, before selecting the candidate roadside point, the region of interest is first selected based on the absolute values ​​|x| and |y| of the x and y coordinates of the point.

[0025] Furthermore, the specific method of step S3 is as follows:

[0026] The roadside feature information in step S3 includes the direction feature, height feature and height change feature of the roadside;

[0027] For the point cloud data of each ring in the region of interest, the starting points are selected in the four quadrants of the three-dimensional coordinate system to calculate the above three features, and compared with the set thresholds to select candidate points on each ring.

[0028] Furthermore, in step S3, the selected candidate points must simultaneously meet the following conditions:

[0029] According to the directional characteristics, for each ring of point cloud data in the area of ​​interest, the starting point of the search is selected in each quadrant, and the N w The covariance matrix and eigenvalue of the points, if the largest eigenvalue D0 and the second largest eigenvalue D1 meet the conditions:

[0030] D0>λD1, then the eigenvector corresponding to D0 represents this N w The direction of each point, combined with the characteristic that the AGV's driving direction is always parallel to the road surface, the first condition for selecting candidate points is:

[0031]

[0032] Among them, (x0, y0, z0) is the feature vector corresponding to D0, θ represents the angle between the feature vector and the unit vector of the x-axis of the sensor coordinate system, and ε is the threshold for candidate point selection;

[0033] According to the height characteristics, N is calculated w The average height of the points The second condition for candidate point selection is:

[0034]

[0035] Among them, h r is the road height, h t is the height threshold, and points exceeding this value are not considered as roadside points;

[0036] According to the height variation characteristics, due to the constraints of the curb geometry, the curb point changes very little in the y direction, while the height in the z direction increases continuously, but there will be no sudden height changes compared to the road surface. This condition is expressed as:

[0037]

[0038] Among them, ν y 、ν z Yes N w The variance of the point in the y direction and the z direction, is the set variance threshold;

[0039] If the above three characteristic conditions are met, then the N of this group w The points are candidate roadside points.

[0040] Furthermore, the specific method of step S4 is: using the RANSAC algorithm twice for the candidate points on the left and right sides of the vehicle sensor respectively, and then using the least squares fitting method to fit the obtained roadside points, and selecting the model with the smallest error.

[0041] Compared with the prior art, the roadside detection method based on multi-line laser radar described in the present invention has the following beneficial effects:

[0042] The curb detection method based on multi-line laser radar described in the present invention can accurately extract curb information in the environment. Compared with the image-based processing method, the path detection method designed in the present invention greatly reduces the impact of environmental factors such as light and weather changes on the detection effect. It has high robustness and low computational complexity, and can extract curb information in the environment in real time. It also fully considers the spatial characteristics of the curb, such as directional characteristics, height characteristics, and height change characteristics. The detection accuracy is high and no missed detection will occur. The extracted curb information can be used for local trajectory planning, edge travel, and other environments in outdoor navigation of AGV. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0044] Figure 1 This is a flow chart of a roadside detection method based on a multi-line laser radar according to an embodiment of the present invention;

[0045] Figure 2 The coordinate diagram of the original point cloud data described in the embodiment of the present invention;

[0046] Figure 3 Extracting a ground coordinate map according to an embodiment of the present invention;

[0047] Figure 4 A schematic diagram of a selected region of interest according to an embodiment of the present invention;

[0048] Figure 5 This is a schematic diagram of a road surface area scanning model according to an embodiment of the present invention;

[0049] Figure 6 This is a schematic diagram of candidate point selection according to an embodiment of the present invention;

[0050] Figure 7 This is a schematic diagram of a fitted curb model according to an embodiment of the present invention;

[0051] Figure 8 This is a result diagram of detecting a roadside point according to an embodiment of the present invention;

[0052] Figure 9 This is a graph of the curb fitting results described in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0054] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0055] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0056] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0057] This method can be applied to an AGV or robot equipped with a 3D laser radar. When the method is applied to an AGV, the method is executed by the vehicle's processor. As an optional method, the embodiment of the present invention selects a Velodyne-16 laser radar as the sensor.

[0058] See also Figure 1 As shown, a roadside detection method based on a multi-line laser radar includes the following steps:

[0059] S1. Scan the vehicle's surroundings using a 3D laser radar. The laser beam emitted by the 3D laser radar will be reflected after hitting an object. After receiving the returned laser beam, the receiver calculates the obstacle coordinates in polar coordinates based on the return time and the propagation speed of the laser. The obstacle coordinates are expressed in polar coordinates to form raw point cloud data, where the raw point cloud data includes the distance value of the point and the angle with the x, y, and z coordinate axes. The points expressed in the polar coordinate system are as follows: Figure 2 As shown;

[0060] S21. Establish a three-dimensional coordinate system with the center of the laser radar as the origin. The vehicle's forward direction is the positive direction of the x-axis, the left of the vehicle is the positive direction of the y-axis, and the top is the positive direction of the z-axis. In order to facilitate the subsequent processing of the point cloud, the point cloud represented by the polar coordinate system is first converted to the Cartesian coordinate system according to formula (1):

[0061]

[0062] Where R is the distance information of point i, γ i is the angle in the vertical direction, θ i is the angle in the horizontal direction, x i 、y i 、z i is the coordinate value of point i converted to the Cartesian coordinate system.

[0063] According to the three-dimensional coordinates of the point cloud and the intrinsic parameters of the sensor, calculate the vertical angle of the point: Angle from the horizontal Then calculate the ring number of each point in the point cloud data and the sequence value of each point on the ring, which serves as the row value and column value of the index point:

[0064]

[0065] For the Velodyne-16 laser radar, 16 laser beams are emitted in each scan, corresponding to 16 rings of return data. Each laser beam contains H = 1600 points, and the minimum vertical scanning angle γ b =15°, vertical angular resolution γ res =2°, horizontal angular resolution θ res =0.2°, the distance from the point to the sensor is:

[0066] S22. Perform ground segmentation on the original point cloud data processed in S21. According to the scanning characteristics of the 3D laser radar, the laser beam with an angle of -15° to 0° in the vertical direction will scan the ground. After the laser scans the road surface, the angle between two adjacent laser beams is significantly smaller than when scanning an obstacle. Figure 3 As shown in the figure, p1 and p2 represent the points when the laser radar scans the ground, and p3 and p4 represent the points when the laser radar scans the obstacle. The condition can be expressed by the following formula:

[0067]

[0068] Where Δx, Δy, and Δz are the differences in the x, y, and z coordinates of corresponding points on adjacent rings. When the angle α between two points in the vertical direction is less than 15°, they are considered to be ground points.

[0069] For non-ground point clouds, the region growing method is used for clustering; first specify a search starting point p s , then compare the distances of the four points in the ps neighborhood with the point respectively, and the largest distance is recorded as r max , the minimum distance is recorded as r mi n, if the specified neighbor point is in p s Calculate the angle in the horizontal direction of the point And if the specified neighbor point is in p s Calculate the angle between two points in the vertical direction of the point Represent the starting point p s The angle between the x-direction and the y-direction of the specified neighboring point in the laser radar coordinate system, and the threshold θ is set Threshold =60°, if the angle is greater than this value, it is considered that the adjacent point and point p s If the neighboring point is of the same type, the search continues with the neighboring point as the center point until there is no point that meets the conditions.

[0070] S23. After segmenting the road point cloud in S22, in order to reduce the amount of calculation, a region of interest is selected, and roadside candidate points are extracted from the ground points within the region of interest. Points within the range of |x| < 30 m, |y| < 7 m are selected as the region of interest. Figure 4 Shown is a top-down view of the region of interest in a frame of ground point cloud.

[0071] S3: Divide the laser radar scanning on the ground into three parts: road surface, roadside and non-road surface area, such as Figure 5 The figure below shows a schematic diagram of the road surface area captured by a ring of data directly ahead in the sensor coordinate system. In the returned laser points, the road surface is a low-lying plane, the curb is a point with continuously increasing height, and the non-road surface area has a certain height increase compared to the road surface. Therefore, the spatial characteristics of the curb are summarized as follows:

[0072] ① Directional characteristics: The direction of the curb is generally parallel to the road surface and in most cases parallel to the direction of vehicle travel;

[0073] ② Height characteristics: The curb is a fixed distance higher than the road surface, generally 10-15 cm, and the height of the curb is always fixed in the same area;

[0074] ③Height change characteristics: The curb increases continuously in height and does not experience height jumps like the surrounding buildings.

[0075] On the xy plane, the reflection points of the road edges on all rings form two line segments, while on the xz plane, the road edges on each ring are some parallel small line segments. The data received at the current time t contains N s Laser beam: G t is the currently extracted ground point, N s is the number of beams of interest.

[0076] The points in the region of interest are processed ring by ring, and each point is divided into P according to whether it is on the left or right side of the sensor coordinate system. l and P r Part, P l is the ground point on the left side of the vehicle, P r is the ground point on the right side of the vehicle, and then according to the formula The horizontal angle of each point is calculated and the P l and P r The horizontal angles of the points in the quadrant are sorted, and the points corresponding to the maximum horizontal angle and the minimum horizontal angle are selected as the starting points for extracting candidate points. The starting points in the four quadrants are named S1, S2, S3, and S4, respectively. Figure 6 As shown in the figure, the dotted lines represent ground points. Starting from points S1, S2, S3, and S4, select adjacent N points according to the direction of the arrows in the figure. w A point is a set of points to determine whether it is a candidate point.

[0077] Standard 1: Principal Component Analysis (PCA) is a method for reducing the dimensionality of data. In this invention, it is used to represent the local directional characteristics of the roadside. PCA is used to calculate N w The principal components of the ground points represent the local features of these points. Starting from the starting point, calculate the adjacent N w The covariance matrix of the point, the coordinates of each point can be regarded as a three-dimensional vector, and the covariance matrix C can be calculated as follows:

[0078]

[0079] Where Cov(x,x), Cov(y,y), and Cov(z,z) are the variances of x, y, and z respectively. Cov(x,y), Cov(y,z), and Cov(x,z) are the covariances of (x,y), (y,z), and (x,z) respectively. Cov(x,y) can be obtained by the following formula:

[0080]

[0081] in and Yes N w The mean of points x and y. According to the spatial characteristics of the roadside ①, the points reflected by the laser radar on the roadside form a small line segment on each ring of data. The eigenvector V corresponding to the maximum eigenvalue D0 of the covariance matrix C can represent this N wThe linear direction of the point. If the largest eigenvalue and the second largest eigenvalue satisfy D0>λD1, it indicates that the more dispersed the group of points is, the stronger the linear relationship is. The eigenvector V corresponding to D0 is used as the first criterion for extracting candidate points:

[0082]

[0083] Where (x0, y0, z0) is the feature vector corresponding to D0, θ represents the angle between the feature vector V and the unit vector of the x-axis of the sensor coordinate system, ε is the threshold for candidate point selection, in this embodiment ε = 30°, N w =15.

[0084] Standard 2: Taking the maximum horizontal angle and the minimum horizontal angle as the starting point, the extraction process is guaranteed to start from the road surface. The height of the starting point is regarded as the road surface height. w Average height of points Used for comparison with the road surface height. Since the curb is 10-15cm higher than the road surface, the correct curb candidate point should satisfy the following formula:

[0085]

[0086] where h r is the road height, h t It is the height threshold, indicating that the value greater than this value is definitely not a roadside point. In the embodiment, the height threshold is set to h t =h r +0.25m.

[0087] Criterion 3: The points returned by the LiDAR from weeds and surrounding buildings in the environment may have similar directional characteristics as the roadside. However, unlike the roadside, the z-values ​​of the buildings are much different from those of the ground points, while the points returned by weeds are more scattered. In order to distinguish the roadside points from the points with similar directional characteristics in the environment, N is introduced. w Variance characteristics of the points. Due to the constraints of the curb geometry, the curb points change very little in the y direction, while the height in the z direction increases continuously, but there is no sudden change in height compared to the road surface. This condition can be expressed as:

[0088]

[0089] Among them, ν y 、ν z Yes N w The variance of the point in the y direction and the z direction, is the variance threshold set, which is set in the embodiment

[0090] If the above three criteria are met, then the N w The points are candidate roadside points.

[0091] S4. After extracting the candidate points of the roadside in each ring data, the next step is to fit the candidate points into a line segment model. Since the candidate points are selected in Pl and Pr respectively, the candidate points in Pl are recorded as Ca l , the candidate point in Pr is recorded as Ca r , in order to solve the problem that RANSAC can only estimate one model, l and Ca r The curb models on the left and right sides of the vehicle are estimated separately.

[0092] In Ca l or Ca r Randomly select two points to generate a straight line model, and then calculate the distance from other points in the candidate point set to the model, such as Figure 7 As shown in the figure, the x-axis represents the front of the vehicle, the y-axis represents the left of the vehicle, and the straight line represents the roadside model to be fitted. When fitting the roadside model, only the two-dimensional information of the point is considered. The algebraic expression of the model is: Ax+By+C=0. Assume that the coordinates of a candidate point are (x i ,y i ), the distance from the point to the model is calculated as follows:

[0093]

[0094] If the distance is less than the distance threshold dc, the point is considered to be a roadside point, otherwise it is considered to be a false detection point. At the same time, the number of roadside points is counted. If the number of roadside points exceeds N, the model is considered reasonable.

[0095] This method uses the RANSAC algorithm to fit candidate points twice: the first time, a smaller N and a larger dc are selected; the second time, based on the first, a higher threshold is selected to obtain a more accurate model. Furthermore, the RANSAC algorithm is combined with the least squares algorithm. After obtaining a model that meets the requirements, the least squares method is used to fit the inliers again, and the model with the smallest error is selected. If the current model does not meet the above conditions, two new points are extracted to construct a straight line model, and the above process is repeated, with the model with the largest number of inliers being the final result.

[0096] The first RANSAC iteration selected a looser condition, the distance threshold d t =0.10, the inlier ratio is set to t=0.4; based on the inlier set of the first fitting, a higher threshold is selected in the second RANSAC iteration, and the distance threshold is d t = 0.05, and the interior point ratio is set to t = 0.8. Figure 8 The figure shows the detection result, and the bold points are the detected roadside points; Figure 9The figure shows a schematic diagram of the fitted curb model. The two line segments in the middle are the curbs, and the curbs on the road are accurately detected.

[0097] Based on the above embodiments, it can be seen that the present invention provides a roadside detection and fitting method based on three-dimensional lidar, which includes a preprocessing method of the original point cloud of the three-dimensional lidar, a comprehensive analysis of the characteristics of the roadside in terms of direction, height and continuity, and the design of corresponding standards, including PCA-based point cloud local feature representation, selection of candidate points of the roadside, and a roadside fitting method, which has high detection accuracy, is not affected by light conditions, has high real-time performance, and can stably and accurately detect roadside points and fit them into line segment models.

[0098] Professionals in this field will appreciate that the steps of the methods and algorithms described in the embodiments disclosed herein can be completed by hardware or hardware related to program instructions, and the program can be stored in random access memory (RAM), internal memory, read-only memory (ROM), registers, hard disk, disk or any other form of storage medium known in the technical field.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A roadside detection method based on multi-line laser radar, characterized in that: The steps include: S1. Obtaining original point cloud data: Scanning the vehicle's surroundings with a 3D laser radar to obtain the coordinates of the original point cloud data; S2. Point cloud data preprocessing: cluster and segment the acquired point cloud data, extract ground point cloud data, select the region of interest, and set the corresponding threshold; S3. Selecting candidate roadside points: Calculating roadside feature information from the point cloud data within the region of interest, comparing it with a set threshold, and selecting candidate roadside points; The specific method of step S3 is as follows: The roadside feature information in step S3 includes the direction feature, height feature and height change feature of the roadside; For the point cloud data of each ring in the region of interest, the starting points are selected in the four quadrants of the three-dimensional coordinate system to calculate the above three features, and then compared with the set thresholds to select candidate points on each ring; In step S3, the selected candidate points must meet the following conditions at the same time: According to the directional characteristics, for each ring of point cloud data in the area of ​​interest, the starting point of the search is selected in each quadrant, and the N w The covariance matrix and eigenvalue of the points, if the largest eigenvalue D0 and the second largest eigenvalue D1 meet the conditions: D0>λD1, then the eigenvector corresponding to D0 represents this N w The direction of each point, combined with the characteristic that the AGV's driving direction is always parallel to the road surface, the first condition for selecting candidate points is: Among them, (x0, y0, z0) is the feature vector corresponding to D0, θ represents the angle between the feature vector and the unit vector of the x-axis of the sensor coordinate system, and ε is the threshold for candidate point selection; According to the height characteristics, N is calculated w The average height of the points The second condition for candidate point selection is: Among them, h r is the road height, h t is the height threshold, and points exceeding this value are not considered as roadside points; According to the height variation characteristics, due to the constraints of the curb geometry, the curb point changes very little in the y direction, while the height in the z direction increases continuously, but there will be no sudden height changes compared to the road surface. This condition is expressed as: Among them, ν y 、ν z Yes N w The variance of the point in the y direction and the z direction, is the set variance threshold; If the above three characteristic conditions are met, then the N of this group w points are candidate roadside points; S4. Fitting of roadside points: performing RANSAC algorithm fitting and least squares fitting on the selected roadside candidate points to obtain a roadside model.

2. The roadside detection method based on multi-line laser radar according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21, point cloud projection; establishing a three-dimensional coordinate system with the three-dimensional laser radar as the origin, converting the original point cloud data represented by polar coordinates into point cloud data in the three-dimensional coordinate system, and projecting the point cloud data into an image containing distances. The image includes row and column values ​​of the point cloud and distance information, where the row value indicates the ring number where the laser point is located, and the column value indicates the number of points on the ring; S22, extracting the ground; clustering and segmenting the point cloud data, where clustering treats points with similar features as points of the same type, and segmentation divides a frame of point cloud into ground points and non-ground points; S23. Select an area of ​​interest; in the ground point cloud data, select an area of ​​interest and set a corresponding threshold.

3. The roadside detection method based on multi-line laser radar according to claim 2, characterized in that: Step S21 specifically includes the following method: A three-dimensional coordinate system is established with the center of the three-dimensional laser radar as the origin, the front of the vehicle as the positive direction of the x-axis, the left of the vehicle as the positive direction of the y-axis, and the top of the vehicle as the positive direction of the z-axis. The point cloud represented by the polar coordinate system is converted into a point cloud in the three-dimensional coordinate system according to formula (1): Where R is the distance information of point i, γ i is the angle in the vertical direction, θ i is the angle in the horizontal direction, x i 、y i 、z i is the coordinate value of point i converted to the Cartesian coordinate system; Calculate the number of rings that each point in the point cloud data is in, and the sequence value of each point on the ring where it is located, as the row value row and column value column of the index point, respectively calculated by formula (2): in, γ i represents the scanning angle of point i in the vertical direction, x i ,y i , z i is the coordinate value of point i; γ b is the lowest vertical scanning angle of the sensor, γ res is the angular resolution of the sensor in the vertical direction; θ i Indicates the angle of the point in the horizontal direction, θ res It is the horizontal angle resolution, H represents the number of points on each laser beam, and the distance information of the points is obtained by calculate.

4. The roadside detection method based on multi-line laser radar according to claim 3, characterized in that: The specific method of step S22 is: cluster the point cloud using the region growing method, specify a point as the search starting point, calculate the angle between the points in the x direction or y direction in its neighborhood and the point, and set the threshold θ Threshold =60°, if the angle is greater than this value, it is considered that the neighboring point and the point are of the same type, and the search continues with the neighboring point as the center point until there is no point that meets the conditions; according to the feature that the angle between the points corresponding to the two adjacent rings on the ground scanned by the laser radar is small in the vertical direction, the feature value is used. Segment ground points, where Δx, Δy, and Δz are the differences in the x, y, and z coordinates of corresponding points on adjacent rings. When the angle α is less than 15°, the point is considered a ground point.

5. The roadside detection method based on multi-line laser radar according to claim 4, characterized in that: In step S23, before selecting a candidate roadside point, firstly, a region of interest is selected based on the absolute values ​​|x| and |y| of the x and y coordinates of the point.

6. The roadside detection method based on multi-line laser radar according to claim 3, characterized in that: The specific method of step S4 is: use the RANSAC algorithm twice on the candidate points on the left and right sides of the vehicle sensor respectively, and then use the least squares fitting method to fit the obtained roadside points, and select the model with the smallest error.

Images