Real time 3D reconstruction of non-structured domestic environment for obstacle avoidance using multiple RGB-D cameras

Abstract

This thesis presents how to obtain a real-time 3D reconstruction of a non-structured scene using multiple RGB-D cameras with the aim to use it for robot planning. Several techniques have been developed, starting from raycasting. In order to obtain better speeds some point filtering procedures have been used. Finally, a temporal filtering technique is used to replace the raycasting approach, and to provide a better reconstruction, a sparse points filter is implemented.