Robustness of 3D Pose Estimation Against Turbidity Using Dual-Eye Cameras and Active/Lighting 3D Marker for Visual-Servoing Based AUV

摘要

This paper presents the 3D pose estimation against turbidity under dark environment by using dual-eye cameras and an active -meaning light emitting- 3D marker for visual-servoing based underwater vehicle. The authors have proposed a 3D-perception based move on sensing (3D-MoS) system using a 3D position and orientation (pose) estimation method with dual-eye cameras that exploits the parallactic nature that enables reliable 3D pose estimation in real-time, named as “Real-time Multi-step Genetic Algorithm (RM-GA).” The active/lighting 3D marker was designed and constructed to improve the 3D pose estimation especially in turbidity and low illumination. In real-time pose estimation, not only recognition but also robustness are important. This paper focus on the robustness of 3D pose recognition performance using dual-eye cameras and 3D marker against turbidity. The experimental results have confirmed that the effectiveness and robustness of the proposed system for the real-time 3D pose estimation under turbidity and night condition.