A fast solution to the approximation of 3D scattered point data from stereo images using triangular meshes

Meshing the 3D scattered point data is an important task in machine vision since it allows the rendering of the 3D structures on a commodity graphics hardware. In this paper, we develop a fast and adaptive method to approximate an organized set of 3D scattered data points from stereo images by a triangular mesh and which can be applied to systems that require low delay response such as telepresence. The method starts by reducing the number of points through the employment of image adaptive sampling techniques. Then, an initial triangular mesh is constructed using the reduced set of points and later refined to minimize a predefined approximation error. Compared to some state of the art schemes, the proposed method results in a significant reduction in the processing time while preserving the quality of the mesh.