基于平面拟合及残差预测的点云压缩方法

Compression Method for Point Cloud Data Based on Plane Fitting and Residual Prediction

激光雷达点云数据三维表征能力出色,但其庞大的数据量给其存储及传输带来了挑战。基于平面拟合的点云压缩将三维点云转换为二维距离图,并使用空间平面拟合坐标点,在一定误差和精度的条件下能够有效降低单帧点云的空间冗余。针对该方法在复杂场景和高拟合精度情况下非拟合点过多导致压缩性能降低的问题,提出了使用残差预测处理未拟合点数据,进一步提升了压缩率。详细介绍了平面拟合残差预测的工作原理及压缩处理框架,使用对称最近邻均方差和位姿估计,从数据误差和应用精度两个方面评估比较不同压缩算法的性能,验证了残差预测对压缩性能提升的有效性。

LiDAR point cloud data has excellent 3D representation capabilities,but its massive data size poses challenges to its storage and transmission.Point cloud compression based on plane fitting transforms 3D point clouds to 2D distance maps,and uses spatial plane fitting coordinates to effectively reduce spatial redundancy of a single frame point cloud within a certain error and accuracy conditions.However,in complex scenarios and high fitting accuracy situations,too many non-fitting points could lead to reduced compression performance.Therefore,this paper proposes to use residual prediction to process the data of non-fitting points,further improving the compression ratio.This paper details the working principle and compression processing framework of residual prediction for plane fitting.The performance of different compression algorithms is evaluated and compared from the perspectives of data error and application accuracy using symmetric nearest neighbor mean square error and pose estimation,and the effectiveness of residual prediction in improving compression performance is verified.