Real-time feature-based 3D map reconstruction for stereo visual guidance and control of mobile robots in indoor environments

The paper deals with 3D map reconstruction for real-time stereo visual guidance and control of mobile robots in indoor hallway environments. The proposed stereo vision system is capable of reconstructing 3D map based on observed geometrical features without a priori knowledge on the environments. Specifically, detected edge features in binocular images are processed in Hough space for matching correspondence before 3D reconstruction. The generated 3D map can be effectively updated using visual tracking techniques for matched features together with correspondence matching algorithms for newly-observed features. Due to the fact that visual tracking is employed to assist maintaining correspondence information and reduce redundant computation, the proposed system can be operated in real time.