Stereo Vision Based Motion Estimation for Underwater Vehicles

Abstract

Unmanned underwater vehicles (UUVs) are ideal tools to implement underwater monitoring missions. Remotely Operated Vehicles (ROVs) are often used to accomplish periodical inspections of jacket structures of offshore platforms. A stereo vision based method is used to navigate the vehicle to locate itself with the aid of environmental reference information. In this paper, a robust scheme is proposed for the missions. After acquisition of stereo image pairs, 3D information is extracted from them. Specific steps include feature extraction and tracking. The SURF (Speeded Up Robust Features) algorithm is used to achieve real-time tracking performance. Due to the noise impact, there are a certain number of outliers in 3D point clouds. A Coarse-To-Fine (CTF) method is proposed to eliminate them. Resorting to SVD (singular value decomposition) method, which can give a close-form solution of rotation matrix and translation vector of the vehicle, the motion is estimated using the maximal subset of inliers. Preliminary experiments show the feasibility of the scheme.