Research On Geometry Compression For LiDAR Point Cloud

In recent years,the new generation of media represented by point cloud has become popular,which has greatly promoted the innovation of social production modes.Depending on the ability to precisely depict large-scale scenes,Li DAR(Light Detection And Ranging,Li DAR)point cloud has become the most important data which directly improves the performance and robustness of relevant applications.However,the huge amount of data generated by Li DAR brings a great challenge to existing networks where bandwidth and storage resources are limited.Efficient and high-quality Li DAR point cloud compression method is of significant importance to point cloud-related applications.Whereas,the disordered,discrete,and non-uniform distribution of Li DAR point cloud makes it difficult to achieve this goal.In this dissertation,targeting the characteristics of Li DAR point cloud,a geometry coding method based on rate-distortion optimization and a geometry coding method based on regular projection is proposed.Meanwhile,a rate control method is proposed to solve the problem of how to efficiently transmit Li DAR point cloud in a bandwidth-limited scenario.The main research results include:1.A rate-distortion optimized Li DAR point cloud geometry compression method is proposed according to the characteristics of Li DAR point cloud.First,inspired by the G-PCC predictive coding solution,a chain-based prediction structure is proposed by considering the prior knowledge of Li DAR acquisition pattern.In addition,a simplified yet effective geometry prediction method is proposed based on this structure.In case the Li DAR calibration parameters are unavailable,a parameter estimation method is also given to ensure that the chain-based structure can be constructed.To further improve the performance of the proposed framework,the relationship between bitrate and distortion is investigated.A ratedistortion optimization model for residual coding is proposed taking into account the characteristics of residuals.Experimental results show that the proposed algorithm outperforms the G-PCC prediction tree coding mode in coding efficiency and computational complexity on multiple data sets.2.To better compress the non-uniform distributed Li DAR point cloud,a regularized projection-based geometry compression method is proposed.In this dissertation,instead of focusing on designing a specific coding tool,efforts were made toward creating a reliable representation and predictive structure for spinning Li DAR point clouds.Firstly,an investigation is conducted to analyze the underlying reason for the non-uniform distribution.A regularized Li DAR point cloud projection representation is proposed by considering the acquisition pattern and mechanical structure of spinning Li DAR.Based on the proposed representation,a reliable prediction structure is constructed and a simple yet effective geometry prediction strategy is designed to remove the spatial redundancy between points.The experimental results show that,compared with the existing state-of-the-art point cloud coding methods,the proposed method can achieve impressive performance under both lossless and lossy coding conditions.3.To efficiently transmit the Li DAR point cloud in a bandwidth-limited scenario,a rate control method for LIDAR point cloud geometry compression is proposed.In this dissertation,according to the distribution characteristics of LIDAR point cloud and the coding process of G-PCC,the rate-distortion performance is analyzed from acquisition distance,scaling level,and quantization parameters,respectively.The coding parameters are simplified by considering the relationship between each factor.The mathematical model of the relationship between parameters and bitrate is then established.Finally,based on the established models,a rate control method for LIDAR point cloud is proposed.The experimental results show that the proposed rate control algorithm can control the bitrate accurately with satisfactory R-D performance.