阅读说明：本技术 一种多组激光雷达外参标定系统及其方法 (Multi-group laser radar external parameter calibration system and method thereof ) 是由 邓梓枫 陈龙 于 2019-10-25 设计创作，主要内容包括：本发明属于激光雷达技术领域,更具体地,涉及一种多组激光雷达外参标定系统及其方法；包含点云分割模块、地面检测的点云匹配模块、地图构建模块、信息融合模块与路径优化模块；利用激光雷达一段时间内的上下文信息,计算激光雷达前后帧之间的点运动关系,加上IMU提供的位姿去除累计的雷达运动畸变,构建区域内的场景地图,即可完成多组激光雷达外参标定工作。本发明能接受一定程度的雷达视野遮挡、共同视野很小或没有的情况,标定完成后可充分利用多雷达之间的互信息,减少视野盲区、提高对外界环境的探测能力。相较于传统的人工或单帧标定方法,本发明适用性广、易于安装、标定结果的鲁棒性与精准度也有很大提升。(The invention belongs to the technical field of laser radars, and particularly relates to a multi-group laser radar external reference calibration system and a method thereof; the system comprises a point cloud segmentation module, a ground detection point cloud matching module, a map construction module, an information fusion module and a path optimization module; and calculating the point motion relation between the front frame and the rear frame of the laser radar by using the context information of the laser radar within a period of time, and removing accumulated radar motion distortion by adding the pose provided by the IMU to construct a scene map in the area, so that the external reference calibration work of a plurality of groups of laser radars can be completed. The invention can receive the conditions of radar view shielding, small common view or no common view to a certain degree, and can fully utilize the mutual information among multiple radars after the calibration is finished, thereby reducing the view blind area and improving the detection capability of the external environment. Compared with the traditional manual or single-frame calibration method, the method has the advantages of wide applicability, easiness in installation and great improvement on the robustness and accuracy of the calibration result.)

1. The utility model provides a multiunit lidar external reference calibration system which characterized in that includes:

a point cloud segmentation module: the output of the laser radar is three-dimensional laser point cloud which comprises the integral description of the external environment, the point cloud segmentation module segments the three-dimensional laser point cloud into a plurality of categories, noise points are filtered, and the segmented semantic point cloud is output to a map construction module and a point cloud matching module of ground detection;

a point cloud matching module for ground detection: the system is used for constructing constraint based on the ground, so that laser point cloud matching can be carried out on a relatively stable plane, the point cloud matching adopts a semantic ICP (inductively coupled plasma) matching algorithm, a map registration strategy of a current frame and surrounding point clouds is selected, the matching process utilizes the category information of point cloud segmentation, meanwhile, pose information provided by an IMU (inertial measurement Unit) is obtained as constraint to improve matching accuracy and efficiency, 6-degree-of-freedom pose estimation of the registered point clouds is output, and the pose estimation is output to a map construction module;

the map building module: the system comprises a scene map building module, a semantic ICP (inductively coupled plasma) closed-loop detection algorithm, an information fusion module, a point cloud map and a point cloud map database, wherein the scene map building module is used for performing fine-grained registration on a newly added point cloud map and a nearby point cloud map, the registration process adopts a graph optimization mode to optimally build the point cloud map, a matching sequence of characteristic point clouds is stored, the subsequently added point clouds can eliminate accumulated drift errors in the previous period of time through the closed-loop detection algorithm of the semantic ICP, and the optimized point cloud map and the matching sequence of the point clouds, namely a path, are output to the information fusion module;

the information fusion module: a multithreading mode is adopted, a plurality of groups of semantic point cloud maps and characteristic point cloud paths from a map building module are respectively received, and information of a plurality of laser radars is integrated; firstly, solving the transformation relation between a plurality of groups of semantic point cloud maps and paths, registering the semantic point cloud maps to obtain the transformation relation between the point cloud maps, fitting the paths of the characteristic point cloud on the transformation relation by adopting a genetic algorithm, and outputting the calibration results of a plurality of groups of auxiliary radar reaching main radars to a path optimization module;

a path optimization module: the system comprises a radar, a main radar, a genetic algorithm and a characteristic point cloud, wherein the radar is used for acquiring initial values of external parameters, applying the external parameters to initially acquired data, converting a coordinate system of each radar, respectively registering semantic point cloud in a global semantic map of the main radar, acquiring paths of the characteristic point cloud, and fitting the paths by using the genetic algorithm to acquire further external parameter offset; and repeating the iteration for multiple times until the change quantity of the external parameters is smaller than a certain threshold value or reaches the maximum iteration time, and finally outputting multiple groups of external parameters of the laser radar.

2. A multi-group laser radar external reference calibration method is characterized in that the calibration system of claim 1 is applied, and the method specifically comprises the following steps:

s1, fixing laser radars on a vehicle body, arranging an IMU (inertial measurement Unit) in the center of the vehicle body, selecting one of the laser radars as a main radar, setting other laser radars as auxiliary radars, and calibrating external parameters of the main radar and the IMU, wherein the position relation among a plurality of groups of laser radars belongs to rigid body transformation;

s2, selecting an indoor and outdoor appropriate environment, and collecting data of a plurality of groups of laser radars and IMUs;

s3, starting a point cloud segmentation module off line, issuing laser point cloud information to a point cloud matching module for ground detection, constructing ground constraint, adopting a semantic ICP (inductively coupled plasma) algorithm, and outputting 6-degree-of-freedom attitude estimation of registered point cloud;

s4, obtaining 6-degree-of-freedom attitude estimation of the semantic point cloud and the registration point cloud, optimally constructing a point cloud map in a map construction module in a graph optimization mode, storing a path of the feature point cloud, adding pose constraints of IMUs to the point cloud map of the main radar, and improving the accuracy of the semantic map of the main radar;

s5, the deviation between the main radar and other radars is roughly measured manually, semantic ICP matching is carried out on the point cloud map of each auxiliary radar and the main radar map by using the initial value of the manual measurement in the information fusion module, the result of registration is used as an initial value, a genetic algorithm is adopted to fit the paths of a plurality of groups of characteristic point clouds, and a preliminary external reference calibration result can be obtained in the step;

s6, respectively carrying out rigid motion transformation on the data of the auxiliary radar by using the calibration result of the step S5, and transferring respective coordinate systems to the main laser radar;

s7, in a semantic point cloud map of the main radar, carrying out respective attitude estimation of point cloud registration on each laser radar to obtain a characteristic point cloud path of each laser radar, and carrying out fine-grained fitting on the path by adopting a genetic algorithm to obtain a more accurate external reference calibration result;

s8, repeating the step S6 and the step S7 until the variation of the external parameters of two times is smaller than a given threshold value or reaches a given iteration number, and finally obtaining the external parameters of a plurality of auxiliary radars reaching the main radar.

3. The method for calibrating the external references of the multiple groups of laser radars according to claim 2, wherein the construction and path optimization of the semantic point cloud map are integrated into the external reference calibration process by using the context information of the laser radars in the calibration process.

4. The method for calibrating the external parameters of the multiple sets of laser radars according to claim 2, wherein all calibration work can be completed indoors and outdoors by only the laser radars and the IMU without depending on the precise global pose provided by the external sensor in the calibration process.

5. The method for calibrating the external reference of multiple sets of lidar according to claim 2, wherein the lidar is a Velodyne 16 line or 32 line lidar or a Robosense 16 line lidar, and the IMU can acquire 3-axis angular velocity and 3-axis acceleration.

Technical Field

The invention belongs to the technical field of laser radars, and particularly relates to a multi-group laser radar external reference calibration system and a method thereof.

Background

In recent years, the continuous progress of laser radar technology provides a good sensor foundation for unmanned automobiles and intelligent mobile robots. By using the laser radar, the distance between the radar main body and the external environment, obstacles and the like can be accurately measured, and the output point cloud can describe the three-dimensional space environment and can be visually displayed. The normal and efficient operation of the equipment cannot be influenced by the laser radar due to the change of illumination intensity and dynamic environment, but in the actual working condition, only a single laser radar is used, so that partial visual field can be lost, and a visual field blind area is generated.

Disclosure of Invention

The invention provides a multi-group laser radar external reference calibration system and a method thereof, aiming at overcoming the defects that the ground environment condition around the vehicle body cannot be detected and the observation result information is sparse by a single laser radar above the vehicle body in the prior art, and the like, and can finish the calibration work of the external reference of the multi-group laser radar and obtain the high-precision calibration result.

In order to solve the technical problems, the invention adopts the technical scheme that: a multi-group laser radar external reference calibration system comprises:

a point cloud segmentation module: the output of the laser radar is three-dimensional laser point cloud which contains the overall description of the external environment, including the road environment around the vehicle body, obstacles and the like; the point cloud segmentation module segments the three-dimensional laser point cloud into a plurality of categories, filters noise points, and outputs the segmented semantic point cloud to the map construction module and the point cloud matching module for ground detection;

a point cloud matching module for ground detection: the system is used for constructing constraint based on the ground, so that laser point cloud matching can be carried out on a relatively stable plane, the point cloud matching adopts a semantic ICP (iterative closest point) matching algorithm, a current frame and surrounding point cloud map registration strategy is selected, the matching process utilizes the category information of point cloud segmentation, meanwhile, the pose information provided by the IMU is obtained as the constraint to improve the matching accuracy and efficiency, 6-degree-of-freedom pose estimation of the registered point cloud is output, and the pose information is output to a map construction module;

the map building module: the system comprises a scene map building module, a semantic ICP (inductively coupled plasma) closed-loop detection algorithm, an information fusion module, a point cloud map and a point cloud map database, wherein the scene map building module is used for performing fine-grained registration on a newly added point cloud map and a nearby point cloud map, the registration process adopts a graph optimization mode to optimally build the point cloud map, a matching sequence of characteristic point clouds is stored, the subsequently added point clouds can eliminate accumulated drift errors in the previous period of time through the closed-loop detection algorithm of the semantic ICP, and the optimized point cloud map and the matching sequence of the point clouds, namely a path, are output to the information fusion module;

the information fusion module: a multithreading mode is adopted, a plurality of groups of semantic point cloud maps and characteristic point cloud paths from a map building module are respectively received, and information of a plurality of laser radars is integrated; firstly, solving the transformation relation between a plurality of groups of semantic point cloud maps and paths, registering the semantic point cloud maps to obtain the transformation relation between the point cloud maps, fitting the paths of the characteristic point cloud on the transformation relation by adopting a genetic algorithm, and outputting the calibration results of a plurality of groups of auxiliary radar reaching main radars to a path optimization module;

a path optimization module: the system comprises a radar, a main radar, a genetic algorithm and a characteristic point cloud, wherein the radar is used for acquiring initial values of external parameters, applying the external parameters to initially acquired data, converting a coordinate system of each radar, respectively registering semantic point cloud in a global semantic map of the main radar, acquiring paths of the characteristic point cloud, and fitting the paths by using the genetic algorithm to acquire further external parameter offset; and repeating the iteration for multiple times until the change quantity of the external parameters is smaller than a certain threshold value or reaches the maximum iteration time, and finally outputting multiple groups of external parameters of the laser radar.

The traditional laser radar point cloud registration method relies on GPS positioning information or vehicle wheel type odometers, wherein the former needs a high-precision GPS and a good outdoor environment, and the latter can cause accumulated errors on the integral of the odometer.

The invention also provides a method for calibrating the external parameters of the multiple groups of laser radars, which comprises the following steps:

s1, fixing laser radars on a vehicle body, arranging an IMU (inertial measurement Unit) in the center of the vehicle body, selecting one of the laser radars as a main radar, setting other laser radars as auxiliary radars, and calibrating external parameters of the main radar and the IMU, wherein the position relation among a plurality of groups of laser radars belongs to rigid body transformation;

s2, selecting an indoor and outdoor appropriate environment, and collecting data of a plurality of groups of laser radars and IMUs;

s3, starting a point cloud segmentation module off line, issuing laser point cloud information to a point cloud matching module for ground detection, constructing ground constraint, adopting a semantic ICP (inductively coupled plasma) algorithm, and outputting 6-degree-of-freedom attitude estimation of registered point cloud;

s4, obtaining 6-degree-of-freedom attitude estimation of the semantic point cloud and the registration point cloud, optimally constructing a point cloud map in a map construction module in a graph optimization mode, storing a path of the feature point cloud, adding pose constraints of IMUs to the point cloud map of the main radar, and improving the accuracy of the semantic map of the main radar;

s5, the deviation between the main radar and other radars is roughly measured manually, semantic ICP matching is carried out on the point cloud map of each auxiliary radar and the main radar map by using the initial value of the manual measurement in the information fusion module, the result of registration is used as an initial value, a genetic algorithm is adopted to fit the paths of a plurality of groups of characteristic point clouds, and a preliminary external reference calibration result can be obtained in the step;

s6, respectively carrying out rigid motion transformation on the data of the auxiliary radar by using the calibration result of the step S5, and transferring respective coordinate systems to the main laser radar;

s7, in a semantic point cloud map of the main radar, carrying out respective attitude estimation of point cloud registration on each laser radar to obtain a characteristic point cloud path of each laser radar, and carrying out fine-grained fitting on the path by adopting a genetic algorithm to obtain a more accurate external reference calibration result;

s8, repeating the step S6 and the step S7 until the variation of the external parameters of two times is smaller than a given threshold value or reaches a given iteration number, and finally obtaining the external parameters of a plurality of auxiliary radars reaching the main radar.

Furthermore, the construction and the path optimization of the semantic point cloud map are integrated into the external reference calibration process by utilizing the context information of the laser radar in the calibration process.

Furthermore, the precise global pose provided by external sensors such as a GPS and the like is not relied on in the calibration process, and all calibration work can be completed indoors and outdoors only by a laser radar and an IMU.

Further, the lidar is a Velodyne 16 line or 32 line lidar or a Robosense 16 line lidar, and the IMU can acquire 3-axis angular velocity and 3-axis acceleration.

Compared with the prior art, the beneficial effects are:

1. the invention does not depend on the accurate global positioning information provided by the outdoor high-precision GPS, only needs the laser radar and a more stable IMU, can finish all calibration work indoors or outdoors, and has wide applicability and low limitation;

2. the invention utilizes the context information of the laser radar for calibration, compared with the traditional calibration method, the method comprises the following steps: based on a manual measurement method or frame information of a radar at a certain moment, the method utilizes more margin and context information, integrates the construction and path optimization of a semantic point cloud map into an external reference calibration process, and can obtain a high-precision calibration result;

3. the calibration method has good robustness, and can still provide a good and accurate calibration result under the conditions of certain radar view shielding and little or no common view.

Drawings

FIG. 1 is a schematic diagram of the calibration system of the present invention.

FIG. 2 is a flow chart of the calibration method of the present invention.

Detailed Description

The drawings are for illustration purposes only and are not to be construed as limiting the invention; for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted. The positional relationships depicted in the drawings are for illustrative purposes only and are not to be construed as limiting the invention.