3D Point Cloud Attribute Compression Based on Cylindrical Projection

Abstract

With the rapid development of 3D acquisition technologies, there are mounting interests in representing real-world objects as 3D point clouds. However, nonuniform point positions and abundant photometry attributes of point cloud (i.e., colors, reflectance and normal directions) result in massive data volume, leading to the heavy overload of transmission. To address this issue, we propose an attribute compression method for static 3D point cloud by taking advantage of 2D video coding technology. The attribute is projected onto 2D images which are generated from cylindrical surface for several times. We require consistent angle for each projection to guarantee the spatial distribution of multiple projected images is similar, then we combine these images into video sequence. 2D video codec is utilized to perform compression. In this way, we benefit from both intra and inter prediction. Objective evaluation shows significant improvement in coding efficiency over reference technologies.