Depth-Perception Based Geometry Compression Method of Dynamic Point Clouds
Chi Chen, Gangyi Jiang, Mei Yu · 2021
Point cloud technologies have been used in many applications, such as virtual reality, automatic driving, and so on. However, huge amount of data in point clouds will seriously affect their application. Thus, how to compress point clouds effectively, especially dynamic point clouds, is an urgent problem to be solved. Based on the video point cloud coding (V-PCC) standard from MPEG, this paper proposes a new depth-perception based geometry compression method of dynamic point clouds. Firstly, with the idea of point cloud's perpendicular projection, a concept of relative depth of point cloud is defined and a scheme is given to calculate the relative depth. At different rotation angles, the depth of point cloud's each point to the observation/projection planes is calculated, and the depth at different rotation angles can be weighted to represent the relative depth of the point cloud. Secondly, based on the relative depth of different points, a point cloud can be divided into four regions, including close shot, medium and close shot, medium and long shot, and long shot. Finally, a rate-distortion optimization strategy based on depth perception is proposed to achieve bitrate control of geometry compression in different regions of the point cloud. Experimental results show that compared with the V-PCC anchor, the proposed method can achieve bitrate saving of 11% for dynamic point cloud's geometry compression, under the premise of ensuring visual quality.