Local-SURF based 3D scene reconstruction for mobile robot

Traditional SURF based 3D reconstruction methods need to extract features on the whole image. Although this technique can provide rich and accurate environment information, however, it is too time consuming and not practical. To solve that problem, a local SURF based 3D reconstruction algorithm for mobile robot is proposed, which includes three main steps: first, the posture variations were captured from posture sensor and milemeter, and then the initial transition matrix was established; second, when finishing extracting features of the previous frame, searching areas were computed from the initial transition matrix and search radius; last, final transition matrix was computed according to the matching points. Experiments showed that the proposed algorithm did improve the speed of scene reconstruction while preserving precision, and moreover, our approach also can correct the posture variation of robot according to the matching results.