Research On LiDAR Point Cloud Geometry Compression Algorithm

With the rapid development of the Light Detection and Ranging(LiDAR)scanning technology,the 3D point cloud collected by LiDAR system can accurately represent 3D scenes and objects.Therefore,LiDAR point cloud is widely used in autonomous driving.However,the massive amount of point cloud data also brings great challenges to the storage and transmission of multimedia systems.How to effectively compress the geometry information of LiDAR point cloud has become a key problem to be solved urgently.At present,the Geometry-based Point Cloud Compression(G-PCC)designed by the Moving Picture Experts Group(MPEG)adopts a method that uses the structural features of lidar to convert geometry coordinates.The predictive geometry coding scheme of radius,azimuth and Laser index information can achieve geometry coding de-redundancy.However,the existing schemes do not fully consider the discontinuity and sparsity of point cloud caused by the accuracy of the acquisition device.Therefore,coding performance can be improved.Aiming at the problem,this thesis propose a predictive geometry coding optimization algorithm based on multi-prediction list and a predictive geometry coding optimization algorithm based on the correlation between multi-beam laser acquisition point,which improves the performance of LiDAR point cloud geometry coding.In this thesis,by analyzing the ratio of the bitstream of each component to be encoded in the total bitstream,it is found that the point with the same Laser index have discontinuous geometry coordinate information due to the sparseness and non-uniform distribution characteristics of point cloud.Aiming at this problem,this thesis analyzes the reasons for the discontinuous geometry information of LiDAR point cloud,and proposes a predictive geometry coding optimization algorithm based on multiple prediction list.Firstly,the algorithm constructs multiple radius prediction lists representing different semantics based on the statistical distribution of the radius information.Secondly,a threshold-based multiprediction list merging algorithm is proposed.Finally,the optimal prediction value is selected based on the rate-distortion optimization principle to achieve efficient prediction of radius information.The experimental results show that,the performance of the predictive geometry coding optimization algorithm based on the multi-prediction list proposed in this thesis can be improved by 1.8% compared with the algorithm of MPEG G-PCC.Based on the structural information of LiDAR,this thesis analyzes the features of the radius of different laser collection point,and proposes an optimization algorithm for predictive geometry coding based on the correlation between multi-beam laser collection point.First,this algorithm uses the radius similarity characteristic of different laser beam collection points under the same rotation azimuth to construct the cross-laser beam prediction point set of radius information,and proposes a selection method of cross-laser beam prediction point.Second,optimal predicted values within and across the laser beam are obtained based on rate-distortion optimization criteria.In addition,this thesis proposes a coding optimization method for the Laser index information and the number of children count for the azimuthbased predictive tree construction method.The experimental results show that,the performance of the predictive geometry coding optimization algorithm based on the correlation between the multi-beam laser acquisition point proposed in this thesis can be improved by 1.8% compared with the algorithm of MPEG G-PCC.