Research On Featare Extraction And Automatic Registration Technology Based On Laser Point Cloud

The development of modern industry and the continuous advancement of artificial intelligence technology have attracted widespread attention of robots and unmanned technologies.In unmanned positioning and navigation systems,laser SLAM(Simultaneous Location and Mapping)and visual SLAM are currently mainly involved in the system,the laser SLAM uses 2D or 3D lidar,2D lidar is mostly used for indoor robot design,and 3D lidar is widely used in the field of unmanned driving.So far,the appearance of lidar is faster and more accurate in measurement,and the information is richer.The object information collected by lidar is a series of scattered points with accurate angle and distance information,called point clouds.Generally,the front end of the laser SLAM system needs to perform continuous frame matching on the laser 3D point cloud data to obtain the robot or vehicle pose.In the process of frame matching,small and dynamic object point clouds are often removed,and large building point clouds are retained.This matching method has higher accuracy in scenes with large buildings,but in the suburbs or lack of large buildings The scene of the object will cause drift phenomenon,resulting in overlapping and distortion of the reconstructed laser point cloud.Aiming at the problem of point cloud frame matching drift in scenes where large buildings are missing or not obvious,this paper proposes a method of registering a single frame laser point cloud to the overall model,using a combination of point-to-point and point-to-line methods.Inter-frame matching.The traditional SLAM system often deletes the tree point cloud information when removing small objects in the point cloud data.The sampling strategy in this article is to use Euclidean when removing small objects such as moving vehicles and pedestrians.Clustering extracts the features of tree point clouds,and uses the features of tree point clouds to perform point cloud matching,which effectively improves the accuracy of point cloud registration.In order to prevent the impact of ground point cloud data on the experiment of removing small objects,ground point cloud segmentation is performed by an improved Random Sampling Consistency(RANSAC)algorithm,which uses normal vectors between sample points parallel to each other and a certain point cloud.The point is parallel to the sample point normal vector as a constraint condition,and the non-curvature curvature is added to zero.It has a good effect on the efficiency of extracting the plane point cloud and the accuracy of extracting the plane point cloud.Li DAR is used to scan the real scene to obtain point cloud data.The center of the selected coordinate system is used as the origin.The normal vector of each point is obtained by the method of least squares,and the scanning point cloud roll,pitch,yaw,and position conversion information are calculated.Use MATLAB to extract the required sampling points,these sampling points contain the required point information.The laser point cloud data that has been located is used as a model,and the traditional ICP algorithm is improved by combining the extracted tree line information.Using the data collected by Velodyne's 16-line Li DAR for matching test without using any loop detection method,the matching results show that the accuracy of the matching algorithm with tree information constraints has been significantly improved and improved.