Polygonal Meshing for 3D Stereo Video Sensor Data

Many visually-guided robotic systems rely on stereo video data streams to obtain surface models of environmental structure. However stereo video-based 3D point clouds are noisier than those produced from laser-based scanners and are subject to areas of sparse point information corresponding to textureless or specular surfaces. This complicates the process of constructing polygonal meshes from these point-clouds. This paper develops an approach to meshing for stereo video surface reconstruction that addresses these issues and, by exploiting the known ego motion of the sensor, obtains surface normal and texture rendering information. This widens the range of applications for which stereo video reconstruction can be applied and opens the possibility of using stereo video to generate models for real-time rendering applications.