Lossless Geometry Coding Method for Dynamic Point Clouds under Octree Structure

To further improve the performance of lossless compression of dynamic point clouds, a method of geometry information entropy coding under an octree structure was proposed. By this method, for intra-frame spatial correlation, the contexts of intra-frame neighbor nodes and the contexts of intra-frame neighbor parent nodes are established based on the coded neighborhood information of the current octree node; for inter-frame temporal correlation, the previously coded point cloud frame is employed as the reference frame, and the octree node inside the reference frame at the same position as the current octree node is used as the reference node. The inter-frame context is modeled using the reference node and its parent node. To fully utilize the modeled contexts and accurately estimate the conditional probability that the current node is non-empty under different contexts, a second-level probability estimation method based on the exponential moving average was proposed. By this method, the probability is first estimated under the context of intra-frame neighbor nodes and the context of intra-frame neighbor parent nodes, respectively. Then, a set of second-level contexts are modeled using the inter-frame contexts and the results of the probability estimation, and the probability is estimated again under the second-level contexts. In the end, the lossless compression is realized by the binary arithmetic encoder. To evaluate the compression performance of the proposed method, the commonly used Microsoft voxelized upper bodies(MVUB)and 8i voxelized full bodies(8iVFB)datasets were selected for performance testing. The experimental results show that the proposed method has a higher lossless compression performance than the methods developed recently, with average coding gains of 2.2% to 28.7%.