TR2020-061

Overhead Reduction in Graph-Based Point Cloud Delivery


Conventional point cloud delivery schemes use graph-based compression to stream three-dimensional (3D) points and the corresponding color attributes over wireless channels for 3D scene reconstructions. However, the graph-based compression requires a significant communication overhead for graph signal decoding, i.e., inverse graph Fourier transform (IGFT), and such large overhead causes a low 3D reconstruction quality due to power and rate losses. We propose a novel scheme of point cloud delivery to significantly reduce the amount of overhead while allowing a small degradation in the 3D reconstruction quality. Specifically, the proposed scheme exploits Givens rotation for the graph-based transform basis matrix to compress the basis matrix into quantized angle parameters. Even when the angle parameters are strongly quantized for compression, the receiver can reconstruct a clean point cloud by using the basis matrix obtained from the quantized angle parameters. Evaluation results show the Givens rotation in the proposed scheme achieves overhead reduction with a slight quality degradation. For example, the proposed scheme achieves 89.8% overhead reduction with 1.3 dB quality degradation compared with the conventional point cloud delivery scheme

 

  • Related Videos