概要

Three-dimensional acquisition is emerging scanning systems raising the problem of efficient encoding of 3D point cloud, in addition to a suitable representation. This paper investigates cloud point compression via a curve-based representation of the point cloud. We aim particularly at active scanning system capable of acquire dense 3D shapes, wherein light patterns (eg bars, lines, grid) are projected. The object surface is then naturally sampled in scan lines, due to the projected structured light. This motivates our choice to first design a curve-based representation, and then to exploit the spatial correlation of the sampled points along the curves through a competition-based predictive encoder that includes different prediction modes. Experimental results demonstrate the effectiveness of the proposed method.