Research On Point Cloud Registration And Mapping On Vehicle-borne Lidar

At present,emergency rescue vehicles have problems such as hysteresis of active suspension mechanisms and poor maneuverability at high speeds,and the actual field environment is complex,dynamic,and uncertain.There is an urgent need for an vehicle-borne preview system that can sense and reconstruct the vehicle’s environment in real time,and feed back environmental information to the suspension system in advance for control.The emergence and development of Li DAR-based measurement technology provides a brand-new technical means for the acquisition of spatial three-dimensional information.The information collected by the Li DAR at different times and positions in the vehicle travel needs to be transferred to a coordinate system and stitched.For this purpose,the registration algorithm and mapping of the vehicle-borne Li DAR point cloud are studied.Combined with the national key research and development plan "Research on the key technology of special chassis and suspension for high mobility emergency rescue vehicles(including fire fighting vehicles)".On the basis of reading a lot of literature,through the registration of the point cloud data collected by the Li DAR during the driving process of the vehicle,and the establishment of a high-precision point cloud map in real time,so as to provide timely environmental information for the vehicle’s active suspension system.The thesis mainly completed the following work:First,analyze the Li DAR and the data characteristics obtained.Introduce the characteristics of Li DAR and its point cloud data,then analyze and describe the characteristics of the point cloud data with normal,curvature and fast feature histogram descriptors,and the preprocessing of point cloud data,such as establishing efficient index model,simplifying,filtering,coordinate transformation,etc.Then,aiming at the disadvantage that ICP algorithm is slow in the process of vehicle-borne lidar application,an ICP registration algorithm based on ground segmentation is proposed.After removing the outliers in each frame,the KD tree method sampling for simplification,set the threshold to eliminate the wrong corresponding point pairs and constraints,and solve the transformation matrix of the point cloud.By comparing several classic registration algorithms through experiments,it is found that the algorithm in this paper can improve the registration speed and accuracy,and can meet the basic requirements of real-time mapping in vehicles.Finally,a hardware time synchronizer is designed to reduce the distortion.Aiming at reducing the accumulated error of point cloud registration after a long time,a system for real-time mapping of 3D point cloud of vehicle-borne Li DAR is designed.The main processes are ground segmentation,feature extraction and matching,and mapping,which can effectively complete the matching and mapping of multi-frame point clouds.It is found through experiments that the algorithm of this system can construct a point cloud map in real time.