Hybrid LIDAR Point Cloud Compression Based on Projection Compensation

Abstract

Lidar point cloud data is being widely used in the fields of autonomous driving, virtual reality, 3D reconstruction, etc., but its huge data volume poses some challenges for its storage and transmission. Compression of point cloud data is necessary. Existing point cloud compression methods have shown different superiorities in compression rate, compression time, compression loss, and other compression-related properties. One of the methods, based on plane fitting, is able to achieve fast compression of point cloud data, but at the cost of large information loss. To address this, we propose a projection-compensated approach that can reduce the information loss of the point cloud while achieving fast compression. In addition, we use a projection method for non-planar fitted points, which further reduces the information redundancy and improves the compression effect. Compared with plane fitting and commonly used octree and prediction tree based point cloud compression methods, our method is close to the optimal octree method in terms of direct information loss, and is better than the prediction tree and plane fitting methods, and the proposed method outperforms the other three methods in terms of application accuracy and compression time with the same compression rate.