Research On Constructing Point Cloud Map Of Enclosed Environment Based On 3D Lidar

3D point cloud map plays an important role in intelligent vehicle perception environment,high-precision positioning,and path planning.Autopilot can achieve a higher accuracy aided location and improve the safety of automatic driving according to the prior environmental information provided by 3D point cloud map.The inertial measurement unit can measure the angular velocity and acceleration of the motion.Therefore,the IMU data can be used to constrain the position and posture of the point cloud and improve the effectiveness of the drawing.In addition,a closed environment can provide closed-loop information of vehicle motion.The combination of the two methods can improve point cloud map accuracy.The paper systematically studies the preprocessing of 3D point cloud data,integrates the front-end point cloud matching algorithm of IMU,and optimizes the pose map based on loop detection and IMU pre-integration,and carries out the experimental verification of point cloud map construction.The main research contents are as follows:(1)The principle of point cloud acquisition of lidar is analyzed.Outliers caused by obstacles and line of sight are filtered.Voxel grid filter is used to sample down to reduce data quantity.The velocity information obtained by IMU is used to eliminate the motion distortion of each point cloud and to synchronize the sensor data.(2)The pre-processing point cloud data is matched among frames to obtain the laser odometer information.Meanwhile,the image construction effect of iterative closest point matching,feature point matching,and normal distribution matching algorithm based on standard optimization technology is compared.IMU data is calculated,which provides a good initial attitude for matching between points cloud frames,and improves the accuracy and matching speed of the laser odometer.(3)The key frame pose of the laser odometer is selected as the vertex to be optimized by introducing the graph optimization algorithm.The constraint relation of position and pose transformation,pre-integration,and closed-loop constraint are added to the adjacent keyframes,the pose map is established,the optimization strategy is formulated,the pose map is modified by nonlinear optimization method,and all keyframe point clouds are transferred to the initial pose coordinate system by using the optimized pose,and the optimization is used to make the optimization of the keyframe point cloud under the initial pose coordinate system The point cloud map is divided by octree grid.(4)The point cloud acquisition vehicle experimental platform is built by Velodyne,IMU,and other sensors.The point cloud data is collected on campus,using campus scene data and open source data to verify the mapping effect of the algorithm.The experimental results show that the 3D point cloud map construction algorithm can reduce the accumulated error of the laser odometer and improve the accuracy of the point cloud map.The results show that the velocity information obtained by solving IMU data can effectively remove the point cloud distortion.The effectiveness of point cloud data can be improved by removing outliers and down sampling.The initial attitude obtained by solving IMU information can improve the matching accuracy of the point cloud.The back end optimizes the position and poses of the keyframe by using IMU pre-integration,the position and attitude relationship of adjacent keyframes,and the closed-loop constraints of detection,which reduces the accumulated error of the laser odometer and makes the drawing effect good.The paper can establish a good campus environment 3D point cloud map,which lays a foundation for the following location based on the point cloud map.