A laser point cloud lane line detection method

Abstract

The application discloses a laser point cloud lane line detection method; high-precision RTK data and laser point cloud data are fused in multiple frames, an optimal curve is fitted in local point cloud and inner and outer points are judged through a prior condition and a set threshold value, an optimal model is fitted through multiple iterations, and multiple lane lines are acquired, so that lane line features are more fine, and detection effect is better and more robust.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic driving, in particular to a laser point cloud lane line detection method. BACKGROUND

[0002] In the field of intelligent driving, lane line detection as a key technology is an important condition to ensure the safe driving of vehicles within a certain range. Lane line detection is usually based on camera sensor method, which has poor detection effect in bad conditions such as night, rain and fog; while laser radar sensor has the advantages of accurate detection and is not affected by environmental light. In the actual road, the lane line also has the situation of defect or discontinuity, and it is difficult to obtain accurate and robust detection effect only by relying on single frame laser radar point cloud data.

[0003] For example, a lane line detection method based on fusion of depth radar point cloud and image data is disclosed in Chinese patent literature, with publication number CN114037968A. The collected RGB image and point cloud data are used to extract and fuse features of the RGB image and point cloud data by using Hourglass network, integrate the fused features by using residual module, and detect lane lines by using the fused features to realize lane line detection by fusing radar point cloud and image. This method combines the advantages of two kinds of data, fuses the two kinds of data under multiple perspectives at the feature layer, uses the difference data between different sensors to complement each other, and increases the robustness and accuracy of lane line recognition. Since this invention analyzes lane lines based on RGB images, it cannot detect lane lines in bad weather. Moreover, in real life, lane lines may be defective, and simply using laser point cloud to detect lane lines cannot accurately obtain the state of lane lines. SUMMARY

[0004] The present application mainly aims at the problems that it is difficult to detect lane lines in bad environment under the prior art, and simple laser radar sensor detection cannot accurately detect lane lines in actual situations; provides a laser point cloud lane line detection method; uses high-precision RTK data and laser point cloud data for multi-frame fusion, fits the optimal curve in the local point cloud and judges the inner and outer points by using the prior condition and the set threshold, and obtains the optimal model by multiple iterations, so as to obtain multiple lane lines, make the lane line features more fine, and improve the detection effect and robustness.

[0005] The above technical problems of the present application are mainly solved by the following technical scheme:

[0006] A laser point cloud lane line detection method, comprising:

[0007] S1, installing a detection device: the inertial navigation system and the laser radar are arranged on the upper part of the vehicle;

[0008] S2, acquiring laser point cloud data: establishing a laser radar coordinate system; obtaining a pose matrix of RTK at each time under the RTK coordinate system as P R =P R t0 , P R t1 ,…, P R tn according to the installation position, obtaining the RTK to laser radar external parameter E RTK→LiDAR ; turning on the laser radar power supply, acquiring the laser point cloud data, and obtaining the point cloud frame under the continuous laser radar coordinate S=S t0 , S t1 ,…, S tn ;

[0009] S3, point cloud preprocessing: preprocessing the point cloud frame, and extracting the candidate lane line points of the point cloud frame;

[0010] S4, multi-frame point cloud fusion: according to the pose matrix P R ti of RTK at t0, t1,…, t n , and the RTK to laser radar external parameter, obtaining the multi-frame local laser radar point cloud S R local under the RTK coordinate;

[0011] S5, RANSAC curve fitting: adopting a quadratic polynomial model to fit the lane line, and obtaining the optimal model, i.e., the fitted curve of the lane line;

[0012] S6, lane line detection: fitting the optimal curve in the local point cloud S R local and obtaining the inlier and outlier points of the curve, filtering the remaining outliers according to the position of the curve in the laser radar coordinate system space and the prior condition, and repeating the operation of step S5 on the filtered points until all lane lines are detected.

[0013] The high-precision RTK data and laser point cloud data are used for multi-frame fusion, the optimal curve is fitted in the local point cloud through the prior condition and the set threshold, and the inlier and outlier points are judged, the optimal model is fitted through multiple iterations, and multiple lane lines are obtained, so that the lane line features are more fine, and the detection effect is better and more robust.

[0014] As preferred, the laser radar coordinate system conforms to the right-hand coordinate system criterion, the x-axis is parallel to the ground plane and points forward, the z-axis is perpendicular to the ground plane and points upward, and the y-axis is perpendicular to the x-z plane and points inward. The coordinate interval is defined, all points are set in the same coordinate system, the x, y and z are used as the marks of the points, the subsequent screening of the inner and outer points after curve fitting is facilitated, and the subsequent point cloud feature extraction is facilitated.

[0015] As preferred, the step S3 specifically comprises: obtaining the local laser radar point cloud S n The point cloud frame at the moment is preprocessed, the detection range of the region of interest is set, the point cloud frame is cropped, and the candidate lane line points of each moment point cloud frame are extracted according to the reflection intensity range of the lane line. The region of interest can be artificially set, and the point cloud frame is cropped and the range is demarcated according to the current demand.

[0016] As preferred, the pose matrix P n at the moment t0, t1,…, t R ti and the extrinsic parameter of the RTK to the laser radar are used to obtain the multi-frame local laser radar point cloud S R local based on the RTK coordinate.

[0017] S R local =∑ n i=0 (P R ti ·E·t i );

[0018] wherein, E=E RTK→LiDAR .

[0019] As preferred, the step S5 specifically comprises: using a quadratic polynomial model to fit the lane line, and combining the RANSAC method to iterate; three points are taken each time to obtain the parameters of the quadratic curve, the distance of each point to the curve model is calculated, if the distance is less than a preset threshold, it is considered as an inner point, otherwise it is an outer point; the number of inner points and the optimal model parameters are recorded; the next cycle is entered, and the optimal model, i.e., the fitted curve of the lane line, is obtained after multiple iterations. The accuracy is gradually increased through multiple iterations, and the outer points are gradually excluded, so that the range is gradually reduced, and the lane line feature is more fine.

[0020] As preferred, the prior condition is that the lane line distance is 3.75 m, and the local point cloud is screened through the prior condition, the points outside the screening range are excluded, the range is gradually reduced, and the fitting accuracy is improved.

[0021] The beneficial effects of the present application are:

[0022] By using high-precision RTK data and laser point cloud data for multi-frame fusion, the point cloud frame data is judged by prior conditions and set thresholds, and the optimal model is fitted through multiple iterations to obtain multiple lane lines, making the lane line features more refined and the detection effect better and more robust. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the installation of an inertial navigation system and a lidar system.

[0024] Figure 2 This is a flowchart of the method used in this invention. Detailed Implementation

[0025] It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0026] The technical solution of the present invention will be further described in detail below through embodiments.

[0027] A laser point cloud lane line detection method includes:

[0028] 1. Installation location diagram:

[0029] like Figure 1 As shown, the inertial navigation system is installed on the roof of the vehicle, and the lidar is installed directly above the windshield, but it is not limited to this installation position. As long as it is installed at a certain height so that the point cloud can scan the lane lines in front, it is acceptable. Assuming that there is no error in the installation of the lidar, a lidar coordinate system is established, which conforms to the right-hand coordinate system rule. The x-axis is parallel to the ground plane and the positive direction is the front of the vehicle, the z-axis is perpendicular to the ground plane and points upward, and the y-axis is perpendicular to the xz plane and points inward.

[0030] High-precision GPS measurements rely on real-time carrier phase differential (RTK) technology, achieving centimeter-level accuracy. Based on the RTK output, the pose matrix in the RTK coordinate system at each moment can be obtained as P. R =P R t0 , P R t1 ,…, P R tn Based on the installation location, the extrinsic parameter E from the RTK to the lidar can be obtained. RTK→LiDAR .

[0031] 2. Obtain point cloud data:

[0032] Turn on the power of the laser radar, get the laser point cloud data through the driver program, each point contains x, y, z coordinates and reflection intensity I, set the laser radar and RTK time synchronization, obtain a set of continuous laser radar coordinate point cloud frames S=S t0 , S t1 ,…, S tn .

[0033] 3. Point cloud preprocessing:

[0034] Preprocess the point cloud frames at t0, t1,…, t n , set the detection range of the region of interest, for example, set x value [0, 100], y value [-15, 15], z value [-3, 3], crop the point cloud frame, extract the candidate lane line points of each time point cloud frame according to the reflection intensity range of the lane line.

[0035] 4. Multi-frame point cloud fusion:

[0036] According to the pose matrix P R ti of RTK at t0, t1,…, t n , and the external parameter of RTK to laser radar, the multi-frame local laser radar point cloud S R local based on RTK coordinates can be obtained: S R local =∑ n i=0 (P R ti ·E·t i ) ; wherein E=E RTK→LiDAR .

[0037] 5. RANSAC curve fitting:

[0038] Get the multi-frame fused local laser point cloud S R local , a quadratic polynomial model can be used for lane line fitting, combined with RANSAC method for iteration, three points are taken each time to calculate the parameters of the quadratic curve, then the distance of each point to the curve model is calculated, if the distance is less than the preset threshold, it is considered as an inner point, otherwise it is an outer point, and the number of inner points and the optimal model parameters are recorded, enter the next loop, finally obtain the optimal model, that is, the fitted curve of the lane line.

[0039] The preset threshold is set by human, the smaller the preset threshold is set, the more accurate the lane line is, and the lane line threshold is usually set to 0.1. However, the preset threshold cannot be infinitely small, and needs to be kept within a certain range, which can be changed according to local regulations.

[0040] 6. Lane line detection:

[0041] According to step 5, an optimal curve can be fitted at the local point cloud S R local

[0042] The optimal curve and its inlier and outlier points are obtained. According to the position of the curve in the laser radar coordinate system space and the prior condition that the lane line spacing is 3.75 m, the remaining outliers are filtered to retain points within the range of 3.75 m, and then step 5 is performed on the filtered points until all lane lines are detected.

Claims

1. A method for lane line detection from a laser point cloud, characterized in that, The method comprises the following steps: S1, installing a detection device: the inertial navigation system and the laser radar are arranged on the upper part of the vehicle; S2, obtain laser point cloud data: establish a laser radar coordinate system; obtain the pose matrix of each time under the RTK coordinate system as P R =P R t0 , P R t1 ,…, P R tn according to the installation position, obtain the RTK to the laser radar external parameter E RTK→LiDAR ; turn on the laser radar power supply, obtain the laser point cloud data, and obtain the point cloud frame under the continuous laser radar coordinate S=S t0 , S t1 ,…,S tn ; S3, point cloud preprocessing: preprocessing the point cloud frame, and extracting candidate lane line points of the point cloud frame; S4. Multi-frame point cloud fusion: Based on RTK data at t0, t1, ..., t... n P pose matrix at time t R ti And the extrinsic parameters from RTK to LiDAR, to obtain the multi-frame local LiDAR point cloud S based on RTK coordinates. R local ; S5, RANSAC curve fitting: adopting a quadratic polynomial model to fit the lane line, and obtaining an optimal model, i.e., a fitted curve of the lane line; S6, Lane Line Detection: In the local point cloud S R local The optimal curve is fitted and its inner and outer points are obtained. Based on the position of this curve in the LiDAR coordinate system space and the prior conditions, the remaining outer points are filtered, and the operation of step S5 is repeated for the filtered points until all lane lines are detected. 2.The method of claim 1, wherein: The laser radar coordinate system conforms to the right-hand coordinate system criterion, the x-axis is parallel to the ground plane and points to the front of the vehicle, the z-axis is perpendicular to the ground plane and points upward, and the y-axis is perpendicular to the x-z plane and points inward.

3. The method of claim 1, wherein: The step S3 specifically comprises: preprocessing the point cloud frames at t0, t1, …, t n The point cloud frames at t0, t1, …, t The point cloud frames at t0, t1, …, t 4.The method of claim 1, wherein: Multi-frame local laser radar point cloud S R local The calculation formula is as follows: S R local =∑ n i=0 (P R ti ·E·t i ); E = E RTK→LiDAR .

5. The method of claim 1, wherein: The step S5 specifically comprises the following steps: adopting a quadratic polynomial model to fit the lane line, and iteratively combining the RANSAC method; three points are taken each time to obtain parameters of a quadratic curve, the distance of each point to the curve model is calculated, if the distance is less than a preset threshold value, the point is considered as an inner point, otherwise, the point is considered as an outer point; the number of inner points and optimal model parameters are recorded; the next cycle is entered, and the optimal model, i.e., the fitted curve of the lane line, is obtained after multiple iterations.

6. The method of claim 1, wherein: The prior condition is that the lane line spacing is 3.75 m.

Images