수중로봇의 3D SLAM을 위한 인공표식의 인식 방법

Abstract

In this paper, we propose an inexpensive and accurate vision based framework of man-made target recognition for autonomous underwater robots. It is known that usual GPS based navigation approaches do not work in the underwater environment. An alternative is topological map-based navigation methods. That is, the robot proceeds SLAM in the area where knowledge of the map of the area is given previously. In such scenario, an important issue is performing exact data association between landmarks in the robots knowledge data base (DB) and observed target objects. In this regards, our contribution lies in implementing a simple vision based approach of the target recognition process in the underwater environment. Our method can classify interesting objects using shape descriptors of each target in the presence of camera motion and background clutters. The proposed method is validated with both synthetic data and real data, and the corresponding results are quantitatively reported in the paper.