Dynamic Point Cloud Denoising via Manifold-to-Manifold Distance

3D dynamic point clouds provide a natural discrete representation of real-world objects or scenes in motion, with wide range applications immersive telepresence, autonomous driving, surveillance, \etc. Nevertheless, are often perturbed by noise due to hardware, software other causes. While plethora methods have been proposed for static cloud denoising, few efforts made the denoising clouds, which is quite challenging irregular sampling patterns both spatially and temporally. In this paper, we represent naturally on spatial-temporal graphs, exploit temporal consistency respect underlying surface (manifold). particular, define manifold-to-manifold distance its counterpart graphs measure variation-based intrinsic between patches domain, provided that graph operators counterparts functionals Riemannian manifolds. Then, construct connectivity corresponding based points adjacent spatial domain. Leveraging initial representation, formulate as joint optimization desired regularized smoothness consistency. We reformulate present an efficient algorithm. Experimental results show method significantly outperforms independent each frame from state-of-the-art approaches, Gaussian simulated LiDAR noise.