전방/하방 카메라 영상 및 인공 표식을 이용한 수중 SLAM

Abstract

Autonomous underwater vehicles (AUVs) are usually equipped with one or more optical cameras to obtain visual data of underwater environments. These cameras can be used to estimate the AUV’s own pose information. In this paper, we propose a vision-based simultaneous localization and mapping (SLAM) of an AUV, where underwater artificial landmarks are designed and used in visual sensing by forward and downward looking cameras. Three types of landmarks are introduced and their detection algorithms are organized in a framework of conventional extended Kalman filter (EKF) SLAM to estimate both robot and landmark states. The proposed method is validated by an experiment performed in an engineering basin. Since DVL suffers from noises in a real ocean environment, we generated synthetic noisy data based on the real sensor data. With this data we verify that the proposed SLAM approach can recover from the erroneous dead reckoning position.