Research On Point Cloud Attribute Compression Algorithm

Three Dimensional（3D）point cloud can represent the 3D scene as well as the contour and structure information of people or objects in the scene through the coordinate information of position,reflectivity,color and other attribute information,which has been widely used in automatic driving,virtual reality and other scenes.However,it is difficult to store and transmit massive point cloud data,so it is of great significance to realize efficient compression of point cloud data.Therefore,Audio Video Standard（AVS）working group published a Point Cloud Reference Software Model（PCRM）.In this model,the spatial correlation of point cloud geometry is used to make differential prediction for the attribute values of each point,so as to remove the spatial redundancy of point cloud attribute information and realize the compression of point cloud data.However,the prediction method in the reference software model does not fully consider the sparsity of lidar point cloud data,which leads to the low accuracy of its attribute value prediction,thus affecting the efficiency of compression.Therefore,based on the PCRM platform,this thesis innovatively considers the distribution characteristics of each point in the point cloud data set dynamically obtained by lidar.According to the sparsity of its distribution,a point cloud attribute compression scheme based on neighbor filtering framework is proposed to replace the original compression method in the reference software model.The proposed scheme aims to achieve accurate utilization of spatial correlation of lidar point cloud,so as to improve the prediction accuracy of attribute values of such point cloud data by the model,and finally realize the improvement of compression efficiency.The research process of this thesis is mainly divided into the following three parts:1.Analyze the distribution characteristics of points in the lidar point cloud,and design the optimization algorithm of point cloud attribute prediction based on neighbor filtering framework with PCRM as the basic technology platform of compression scheme;2.Based on the maximum attribute difference parameter,a neighbor filtering framework is constructed with this parameter as the decision threshold;3.Obtain maximum attribute difference parameters based on Rate-DistortionOptimization（RDO）method.The experiment in this thesis follows Common Test Conditions（CTC）of AVS point cloud compression,and the data set used in the experiment is Cat2-frame test sequence provided by AVS point cloud working group,which contains 5 sequences collected by lidar.Compared with the original method of reference platform,the time complexity of codec is basically unchanged under four kinds of CTC test conditions,and the coding performance is improved under each test condition.Under the condition of the first three types of attributes with different limits of loss,the average coding gain is 0.2%,0.3%,0.4%,respectively,and the highest coding gain of each sequence is 0.5%,0.6%,0.7%.Under the condition of the fourth type of geometric lossless and attribute lossless,the average bit saving rate is 1.2% and the highest bit saving rate is 2.3%,which verifies the effectiveness of the proposed scheme in improving the lidar point cloud compression efficiency.