Research On LiDAR Point Cloud Lossless Geometry Coding Based On Geometric Reconstruction

As demands for 3D information applications are becoming more and more prominent,point cloud is gradually attracting attention as one of the excellent 3D information representation formats.With the rapid development of LiDAR?Light Detection and Ranging,LiDAR for short?technology,the point cloud data collected by LiDAR is widely used in the field of autonomous driving.It used as the main input data for navigation,obstacle avoidance,target tracking and recognition and other tasks.However,due to the enormous volume of point cloud data,transmitting and storing the data requires large bandwidth and storage space.It could be a critical bottleneck,especially in tasks such as autonomous driving.In order to solve the problem of lossless geometry coding of unstructured LiDAR point cloud for autonomous driving,this paper used the Standard point cloud coding software: Point Cloud Coding Exploration Model v0.2?PCEM v0.2?released by the Audio Video coding Standard Workgroup of China?AVS Standard Working Group?as the basic platform.The paper optimized the point cloud lossless geometry coding part of the PCEM according to the spatial geometric characteristics of the LiDAR point cloud.The LiDAR three-dimensional point cloud has the characteristics that the spatial distribution of the point cloud is large,and the point cloud subclusters,such as cars,pedestrians,trees,buildings and so on,are relatively discrete.There is a lot of spatial redundant information being coded in the process of lossless geometry coding.In view of these features,this paper studied the spatial distribution characteristics of unstructured 3D LiDAR point clouds for autonomous driving,and tried removing point cloud spatial redundancy through following methods: two-dimensional mapping,space division,point cloud reconstruction.The article also studied the influence of the above spatial redundancy removing algorithms on the lossless point cloud geometry coding rate.The main research contents are divided into three parts shown as follows:1.Research and summarize the spatial distribution characteristics of unstructured LiDAR 3D point clouds for autonomous driving.2.Explore reasonable point cloud segmentation and reconstruction schemes based on the unstructured LiDAR 3D point cloud distribution characteristics.3.Based on the PCEM platform,design a lossless geometry coding scheme for the reconstructed LiDAR 3D point clouds.The final performance was tested on two datasets,Ford and Livox,which are two standard LiDAR 3D point cloud test datasets for autonomous driving provided by the AVS Workgroup.Compared with the original platform,PCEM,the lossless geometry coding reaches a compression ratio gain up to-1.61% on the dataset called Livox₀2_all_in_one₁mm and-0.71% in average in All Intra?AI?configuration.