Distributed Optimization Based on Graph Filter for Ensuring Invariant Simplification of High-Volume Point Cloud

Abstract

In many consumer electronic applications, three-dimensional (3D) point cloud data (PCD) is commonly generated in data-driven models. The expensive burden of computation and storage would be brought out, as the point number reaches billions in the large-scale augmented or virtual reality models. Based on the concept of graph signal, a new PCD simplification approach is developed to handle the high-volume PCD and develop a feature-controllable mechanism. In particular, PCD simplification is modeled as an optimization problem with the hybridization of contour and planar feature items. To reduce the complexity of computation, invariant decomposition is proposed, and distributed optimization is developed for the feature controllable PCD simplification. The self-adoption scheme is adopted to improve the convergence rate. Verification experiments and applications on different scenarios demonstrate the effectiveness and feasibility of the proposed simplification approach.