Predictive Generalized Graph Fourier Transform for Attribute Compression of Dynamic Point Clouds

As 3D scanning devices and depth sensors advance, dynamic point clouds have attracted increasing attention as a format for objects in motion, with applications various fields such immersive telepresence, navigation autonomous driving gaming. Nevertheless, the tremendous amount of data significantly burden transmission storage. To this end, we propose complete compression framework attributes clouds, focusing on optimal inter-coding. Firstly, derive inter-prediction predictive transform coding assuming Gaussian Markov Random Field model respect to spatio-temporal graph underlying clouds. The proves be Generalized Graph Fourier Transform terms decorrelation. Secondly, refined motion estimation via efficient registration prior inter-prediction, which searches temporal correspondence between adjacent frames irregular Finally, present based inter-coding our previously proposed intra-coding, where determine mode from rate-distortion optimization offline-trained ?-Q model. Experimental results show that achieve around 17% bit rate reduction average over competitive cloud methods.