Research On Segmentation Algorithm Based On 3D Lidar Point Cloud Data

In recent years,with the advancement and development of computer science and 3D scanning technology,the cost of acquiring 3D point cloud data has dropped significantly.Compared with traditional two-dimensional images,three-dimensional point cloud data can more completely express environmental feature information,and has unique advantages in terms of spatial structure.Therefore,the analysis and processing of three-dimensional point cloud data has become a current research hotspot.Among them,point cloud segmentation is the difficulty and hotspot of 3D point cloud processing,and it is also the basis and key to all subsequent research such as target detection,object recognition and tracking algorithms.Realizing effective segmentation of point cloud data has great research value.This paper is oriented to outdoor scenes and takes the three-dimensional point cloud data obtained by lidar as the research object.Considering the real-time and accuracy requirements of the point cloud segmentation algorithm in the outdoor environment,a point cloud segmentation algorithm based on depth map is improved,and a three-dimensional point cloud segmentation algorithm based on region clustering is proposed.Finally,the effectiveness of the algorithms is verified through experiments.The main work of this paper is as follows:1.First,it introduces the current research status of point cloud segmentation algorithms at domestic and foreign,and in-depth study of several important technologies in the pre-processing link before point cloud segmentation,including several key steps such as data collection,data spatial structure establishment,point cloud filtering,and point cloud simplification.The result of point cloud data preprocessing is directly related to the performance of the subsequent segmentation algorithms.2.Aiming at the real-time and accuracy requirements of the point cloud segmentation algorithm in the outdoor environment,a point cloud segmentation algorithm based on depth map is improved.This algorithm converts 3D point cloud data into a 2D depth image structure,avoids directly processing large-scale point cloud data in 3D space,and effectively improves the real-time performance of the algorithm.The algorithm mainly uses the attribute of the inclination angle between adjacent points to judge the clustering of ground points,and uses an adaptive breakpoint detector to complete the clustering and segmentation of non-ground target points.The experimental results show that the improved algorithm has excellent performance in real-time,and at the same time has a lower false segmentation rate.3.Aiming at special factors such as slope and potholes on the actual ground,a point cloud segmentation algorithm based on region clustering is proposed.The algorithm performs segmental fitting to the ground plane in the ground segmentation stage,which effectively solves the mis-segmentation defects of some other ground segmentation algorithms when facing non-ideal ground surface.The two attributes of adaptive Euclidean distance threshold and angle threshold are combined to classify non-ground target points by similarity.By applying different distance thresholds in different scanning areas,the segmentation accuracy of the algorithm is effectively improved.After experimental verification,the effectiveness of this algorithm is proven.