Stereo Visual Odometry for Mobile Robot

Abstract

This paper presents a visual localized approach, based on Speeded Up Robust Features (SURF) from a stereo system with optimization tool constraints to obtain high matching precision between images. The contribution of this paper presents a robust visual odometry and a 3D reconstruction algorithm based on Adaptive Iterative Closest SURF Point (AICSP). This algorithm combines the robustness of SURF to detect and match a good feature, and the accuracy of the Adaptive ICP algorithm, which is used to give more importance for near 3D weighted points with their inverse depth. The proposed algorithm is validated and compared to other optimization techniques based on Singular Values Decomposition (SVD) and Quaternion. Experimental results show robustness, accuracy and acceptable outcomes from our algorithm in both: indoor and outdoor environments using Pioneer 3-AT.