Research On LiDAR Point Cloud Data Compression Algorithm

With the rapid development of artificial intelligence applications such as autonomous driving,3D reconstruction and other related industries,LiDAR has become one of the most representative 3D information acquisition devices in the industry.The massive point cloud data obtained by LiDAR can describe large-scale 3D scenes in great detail,but the huge amount of point cloud data has brought huge pressure on current storage resources and network bandwidth.Therefore,it is of great significance to research the compression algorithm of LiDAR point cloud.However,the existing LiDAR point cloud compression schemes do not fully consider the distribution characteristics of LiDAR point cloud,so they cannot effectively use the spatial correlation of point clouds to remove spatial redundancy,so their compression performance is still not high enough.In order to further improve the compression performance of LiDAR point cloud,this paper proposes a two-dimensional regularized projection algorithm for LiDAR point cloud when the calibration parameters of LiDAR are known.In order to solve the problem that the two-dimensional regularized projection algorithm cannot be effectively applied due to the unknown calibration parameters,this paper further proposes an estimation algorithm for the LiDAR calibration parameters.In order to make full use of the spatial correlation of LiDAR point cloud to remove spatial redundancy and then improve point cloud compression performance,this paper proposes a two-dimensional regularized projection algorithm for LiDAR point cloud.The algorithm starts from the LiDAR point cloud itself,explores the system structure and acquisition characteristics of the LiDAR in detail,and conducts an in-depth analysis of the distribution characteristics of the LiDAR point cloud.When corresponding LiDAR calibration parameters are known,the two-dimensional regularized projection scheme for LiDAR point cloud is further determined based on the derived conversion relationship.This scheme firstly establishes a two-dimensional plane structure for the point cloud,and then projects the point cloud to the plane structure by the two-dimensional plane mapping,and finally a twodimensional plane structure with regular distribution after projection is obtained.The experimental results show that the two-dimensional regularized projection algorithm proposed in this paper is accurate and effective,and can obtain a representation that fully reflects the spatial correlation of point cloud.The point cloud encoding scheme based on this algorithm is compared with G-PCC point cloud encoding schemes that yield an average performance improvement of 10.1% when lossless compression is performed,and an average gain of 3.4% on BD-rate of the point-to-point distortion metric(D1)when lossy compression is performed.Since some LiDAR point cloud currently do not provide corresponding LiDAR calibration parameters,the two-dimensional regularized projection algorithm proposed in this paper cannot be effectively applied.Therefore,this paper further proposes a LiDAR calibration parameters estimation algorithm,which is divided into two parts: the estimation of the vertical calibration parameters and the estimation of the horizontal calibration parameters.The estimation of the vertical calibration parameters uses the built neural network model to classify the point cloud,and then according to the relationship of each class of points in the vertical direction,solves the estimated value of the vertical calibration parameters of LiDAR.The estimation of the horizontal calibration parameters uses the estimated or known vertical calibration parameters to classify point cloud,and deduces the relationship that each class of points satisfies in the horizontal direction,and then selects two points from them and builds a system of equations to solve the estimated value of horizontal calibration parameters of LiDAR.The experimental results show that the proposed LiDAR calibration parameters estimation algorithm is accurate and effective,and can well solve the problem of unknown calibration parameters of LiDAR for the two-dimensional regularized projection algorithm proposed in this paper.