A method for online calibration of external parameters of laser radar

Abstract

The present invention discloses an online calibration method for laser radar external parameters, comprising the following steps: step S1) obtaining point cloud data and preprocessing; step S2) extracting ground information from the point cloud data; step S3) using the RANSAC algorithm, fitting the target ground plane using the ground information; step S4) calculating the transformation matrix using the normal vector coordinates of the target ground plane and the normal vector coordinates of the ideal ground plane; step S5) obtaining roll, pitch and tz according to the transformation matrix; step S6) using a laser odometer, calibrating to obtain yaw. The present invention does not rely on specific scenarios and tools, does not add additional equipment, has simple steps, realizes online real-time updating of laser radar external parameters, corrects the external parameter errors of the laser radar in real time, and has good reliability.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to an online calibration method for external parameters of a laser radar. Background Art

[0002] In recent years, with the rapid development of intelligent driving technology, laser radar has the characteristics of high accuracy, large ranging range, and no light influence. It is widely used in environmental perception fields such as obstacle detection, real-time positioning and map construction of intelligent driving vehicles. As one of the commonly used sensors for autonomous driving vehicles, laser radar needs to convert point cloud data to the vehicle coordinate system through external parameters before it can be used after obtaining it and identifying obstacles. However, during the driving process of the vehicle, the external parameters of the laser radar will temporarily change due to factors such as bumps, and the mounting bracket will also undergo mechanical deformation over time, resulting in permanent changes in the external parameters. These factors will cause the obstacles identified by the laser radar to be unable to be accurately converted to the vehicle coordinate system, thereby affecting the judgment and decision-making of the autonomous driving system. At present, the external parameter estimation of the laser radar is mostly done by offline calibration. The system needs to use specific tools to collect data in specific scenarios, and then calculate the external parameters offline. This method has the following disadvantages: 1) It requires specific scenarios and equipment, and the calibration process is cumbersome and has many restrictions; 2) It can only be calibrated offline, and the external parameter error of the laser radar cannot be corrected online in real time. Summary of the invention

[0003] The present invention is mainly intended to solve the problem that the existing offline laser radar extrinsic parameter calibration method requires specific scenarios and equipment and the calibration process is cumbersome. An online laser radar extrinsic parameter calibration method is provided, which does not rely on specific scenarios and tools and does not add additional equipment. It can realize online real-time updating of laser radar extrinsic parameters and real-time correction of laser radar extrinsic parameter errors.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A laser radar extrinsic parameter online calibration method comprises the following steps: step S1) obtaining point cloud data and preprocessing; step S2) extracting ground information from the point cloud data; step S3) using the RANSAC algorithm, fitting the target ground plane using the ground information; step S4) calculating the transformation matrix using the normal vector coordinates of the target ground plane and the normal vector coordinates of the ideal ground plane; step S5) obtaining roll, pitch and tz according to the transformation matrix; step S6) using the laser odometer to calibrate and obtain yaw. At present, laser radar extrinsic parameter estimation mostly adopts offline calibration. The system needs to use specific tools to collect data in specific scenarios, and then calculate the extrinsic parameters offline. The process is cumbersome, with many restrictions, and can only be calibrated offline, and the laser radar extrinsic parameters cannot be updated online in real time, and there are errors. To this end, the present invention proposes a laser radar extrinsic parameter online calibration method, which does not rely on specific scenarios and tools, does not add additional equipment, and realizes online real-time updating of laser radar extrinsic parameters and corrects extrinsic parameter errors. Since the external parameters between the laser radar and the vehicle body can be represented by the rotation parameters Euler angles roll, pitch, yaw and the translation parameters tx, ty, tz, among which tx and ty can be directly measured and have little influence on the external parameters, the present invention mainly focuses on the calibration of roll, pitch, yaw and tz parameters. The specific process is as follows: firstly, a laser radar is installed on the car, point cloud data is collected by the vehicle-mounted laser radar, and the collected point cloud data is preprocessed; then the preprocessed point cloud data is ground segmented, that is, ground information is extracted from the point cloud data, and the RANSAC algorithm is used to perform an iterative calculation process using the ground information to fit the target ground level. The surface is calculated to obtain the normal vector coordinates of the target ground plane; the ground plane parallel to the chassis of the car is taken as the ideal ground plane, and the normal vectors of the target ground plane and the normal vectors of the ideal ground plane are calculated to obtain the transformation matrix; then the rotation parameters Euler angle roll, pitch and translation parameter tz are extracted from the transformation matrix; finally, the laser odometer is used to calibrate the rotation parameter Euler angle yaw, that is, the laser odometer is used to obtain the relative posture transformation matrix of the car from time t0 to time t1, and according to the relative posture transformation matrix, the point cloud data at time t1 is converted to the t0 coordinate system to obtain (x1, y1), and according to (x1, y1), yaw is calculated. The present invention calibrates the coordinate transformation from the laser radar to the car body on the self-driving car online, thereby obtaining the relative posture of the car body and the laser radar, and then realizing the fusion and calibration of the car body and other on-board sensors.

[0006] Preferably, the specific process of step S6 includes the following steps: step S61) controlling the car to travel in a straight line; step S62) using a laser odometer to obtain a relative posture transformation matrix of the car from time t0 to time t1; step S63) according to the relative posture transformation matrix, converting the point cloud data at time t1 to the t0 coordinate system to obtain (x1, y1); step S64) according to (x1, y1), calculating and obtaining yaw. The present invention controls the car to travel in a straight line, that is, the car does not change on the horizontal plane (ground direction) (the yz plane and the xz plane do not change). When the car travels along a straight line from the position at time t0 to the position at time t1, the point cloud data obtained by the lidar at time t0 is recorded as P0, and the point cloud data obtained at time t1 is recorded as P1. P0 and P1 are aligned through the point cloud matching algorithm to obtain the relative pose transformation matrix of the point cloud data from P0 to P1. According to the relative pose transformation matrix, the point cloud data of P1 is converted to the P0 coordinate system, and the coordinates of a point in the P0 coordinate system are (x1, y1). The yaw angle is calculated based on (x1, y1).

[0007] As a preferred embodiment, the specific process of step S62 is as follows: when the car travels along a straight line from the position at time t0 to the position at time t1, the point cloud data of the car at time t0 is P0, and the point cloud data at time t1 is P1. The point cloud data P0 and the point cloud data P1 are registered by the point cloud matching algorithm to obtain the relative pose transformation matrix from P0 to P1. When the car travels along a straight line from the position at time t0 to the position at time t1, the point cloud data obtained by the laser radar at time t0 is recorded as P0, and the point cloud data obtained at time t1 is recorded as P1. P0 and P1 are registered by the point cloud matching algorithm to obtain the pose transformation of the point cloud data from P0 to P1.

[0008] Preferably, the calculation formula of step S64 is:

[0009]

[0010] Based on the above formula, the yaw angle is calculated according to (x1, y1).

[0011] Preferably, in an ideal scenario, the ideal ground plane is parallel to the plane of the car chassis. In an ideal scenario, the car body is parallel to the ideal ground plane (ground level), that is, the car chassis is parallel to the ideal ground plane. At this time, the laser radar installed in the car body is used to scan the points in space, and the ground points are extracted from these points. If the bottom surface of the radar is parallel to the car plane (that is, the plane of the car chassis), that is, the radar is parallel to the ideal ground plane, the ground points extracted by the radar should coincide with the ideal ground plane, and roll and pitch are both 0. However, there are errors in the actual installation process, the bottom surface of the radar is not parallel to the ideal ground plane, and the target ground plane extracted by the radar does not coincide with the ideal ground plane. Assuming that the normal vector of the ideal ground plane upward is [0,0,1], the normal vector to the right is [0,1,0], and the normal vector of the target ground plane extracted from the laser radar point cloud data is F. By calculating the transformation matrix of F and the normal vector of the ideal ground plane, the roll and pitch angles and tz can be obtained.

[0012] Preferably, tx and ty are obtained by measurement. tx and ty can be directly measured and have little influence on external parameters. The present invention mainly focuses on the calibration of roll, pitch, yaw and tz parameters.

[0013] Therefore, the advantages of the present invention are:

[0014] (1) It does not rely on specific scenarios and tools, does not require additional equipment, and has simple steps;

[0015] (2) Realize online calibration of lidar external parameters and correct the lidar external parameter errors in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of an embodiment of the present invention.

[0017] Figure 2 It is a schematic diagram of the structure of the laser radar, the automobile chassis plane and the ideal ground plane in an embodiment of the present invention.

[0018] Figure 3 Schematic diagram of the structure of the target ground plane and the ideal ground plane in the embodiment of the present invention.

[0019] Figure 4 Schematic diagram of the three-dimensional space coordinates of a car in an embodiment of the present invention.

[0020] Figure 5 2 is a schematic diagram of an xy plane of a car in an embodiment of the present invention.

[0021] 1. LiDAR 2. Ideal ground plane 3. Target ground plane 4. Car chassis plane. DETAILED DESCRIPTION

[0022] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, a laser radar external parameter online calibration method includes the following steps: step S1) obtaining point cloud data and preprocessing; step S2) extracting ground information from the point cloud data; step S3) using the RANSAC algorithm to fit the ground information to obtain the target ground plane; step S4) using the normal vector coordinates of the target ground plane and the normal vector coordinates of the ideal ground plane to calculate the transformation matrix; step S5) according to the transformation matrix, obtaining roll, pitch and tz; step S6) using the laser odometer to calibrate and obtain yaw. Since the external parameters between the laser radar and the vehicle body can be represented by the rotation parameters Euler angles roll, pitch, yaw and the translation parameters tx, ty, tz, among which tx and ty can be directly measured and have little influence on the external parameters, this embodiment mainly focuses on the calibration of roll, pitch, yaw and tz parameters. The specific process is as follows: first, a laser radar is installed on the car, point cloud data is collected by the vehicle-mounted laser radar, and the collected point cloud data is preprocessed; then the preprocessed point cloud data is ground segmented, that is, ground information is extracted from the point cloud data, and the RANSAC algorithm is used to perform an iterative calculation process using the ground information to fit the target ground level. The surface is then used to obtain the normal vector coordinates of the target ground plane; the ground plane parallel to the chassis of the car is taken as the ideal ground plane, and the normal vectors of the target ground plane and the ideal ground plane are calculated to obtain the transformation matrix; then the rotation parameters Euler angle roll, pitch and translation parameter tz are extracted from the transformation matrix; finally, the laser odometer is used to calibrate the rotation parameter Euler angle yaw, that is, the laser odometer is used to obtain the relative posture transformation matrix of the car from time t0 to time t1, and according to the relative posture transformation matrix, the point cloud data at time t1 is converted to the t0 coordinate system to obtain (x1, y1), and yaw is calculated based on (x1, y1).

[0024] The specific process of step S6 includes the following steps: step S61) controlling the car to travel in a straight line; step S62) using the laser odometer to obtain the relative posture transformation matrix of the car from time t0 to time t1; step S63) according to the relative posture transformation matrix, converting the point cloud data at time t1 to the t0 coordinate system to obtain (x1, y1); step S64) according to (x1, y1), calculating yaw. The present invention controls the car to travel in a straight line, that is, the car does not change on the horizontal plane (ground direction) (such as Figure 4As shown in the figure, forward is y, right is x, upward is z, and the yz plane and xz plane do not change). When the car travels along a straight line from the position at time t0 to the position at time t1, the point cloud data obtained by the laser radar at time t0 is recorded as P0, and the point cloud data obtained at time t1 is recorded as P1. P0 and P1 are registered by the point cloud matching algorithm to obtain the relative pose transformation matrix of the point cloud data from P0 to P1; according to the relative pose transformation matrix, the point cloud data of P1 is converted to the P0 coordinate system, and the coordinates of a certain point in the P0 coordinate system are obtained as (x1, y1), as shown in the figure. Figure 5 As shown; according to (x1, y1), according to the formula Calculate the yaw angle.

[0025] In an ideal scenario, the ideal ground plane 2 is parallel to the vehicle chassis plane 4. Figure 2 As shown, in an ideal scenario, the vehicle body is parallel to the ideal ground plane 2 (ground level), that is, the vehicle chassis is parallel to the ideal ground plane 2 ( Figure 2 The square in the figure represents the ideal ground plane 2 and the car chassis plane 4, and the two planes are parallel). At this time, the laser radar 1 installed on the car body scans the points in space and extracts the ground points from these points. If the bottom surface of the radar is parallel to the car plane (i.e., the car chassis plane 4), that is, the radar is parallel to the ideal ground plane 2, then the ground points extracted by the radar should coincide with the ideal ground plane 2, and both roll and pitch are 0. However, there are errors in the actual installation process, and the bottom surface of the radar is not parallel to the ideal ground plane 2, such as Figure 3 As shown, the target ground plane 3 extracted by the radar does not coincide with the ideal ground plane 2 ( Figure 3 The square in represents the target ground plane 3 and the ideal ground plane 2, and the two planes do not overlap). Assume that the upward normal vector of the ideal ground plane 2 is [0,0,1], and the right normal vector is [0,1,0]. The normal vector of the target ground plane 3 extracted from the lidar point cloud data is F. By calculating the transformation matrix between F and the normal vector of the ideal ground plane 2 (i.e. Figure 3 The roll and pitch angles as well as tz can be calculated by using the transformation matrix of the two plane coordinate axes in the image.

Claims

1. A laser radar external parameter online calibration method, characterized in that: The following steps are involved: Step S1: Obtain point cloud data and perform preprocessing; Step S2: extracting ground information from point cloud data; Step S3: Use the RANSAC algorithm to fit the ground information to obtain the target ground plane; Step S4: Calculate a transformation matrix using the normal vector coordinates of the target ground plane and the normal vector coordinates of the ideal ground plane; Step S5: Obtain roll, pitch and tz according to the transformation matrix; Step S6: using the laser odometer to calibrate and obtain yaw; comprising the following steps: Step S61: Control the car to travel in a straight line; Step S62: using the laser odometer to obtain the relative posture transformation matrix of the vehicle from time t0 to time t1; Step S63: According to the relative posture transformation matrix, the point cloud data at time t1 is converted to the coordinate system t0 to obtain (x1, y1); Step S64: Based on (x1, y1), the arc tangent of y1 / x1 is calculated to obtain yaw.

2. The method for online calibration of laser radar external parameters according to claim 1, characterized in that: The specific process of step S62 is: when the car travels along a straight line from the position at time t0 to the position at time t1, the point cloud data of the car at time t0 is P0, and the point cloud data at time t1 is P1. Through the point cloud matching algorithm, the point cloud data P0 and the point cloud data P1 are aligned to obtain the relative posture transformation matrix from P0 to P1.

3. The method for online calibration of laser radar external parameters according to claim 2, characterized in that: The calculation formula of step S64 is:

4. The method for online calibration of laser radar external parameters according to claim 1, characterized in that: In an ideal scenario, the ideal ground plane is parallel to the plane of the vehicle chassis.

5. The method for online calibration of laser radar external parameters according to claim 1, characterized in that: tx and ty are obtained through measurement.

Images