Docking Experiments in Ocean for AUV by 3-dimensional Move on Sensing (3D-MoS) with Dual-eye Cameras

摘要

A dual-eye visual servoing system for underwater vehicle has been developed. Pose estimation based on the 3D model-based recognition using dual-eye cameras and 3D marker was proposed for real-time pose tracking. The Multi-step Genetic Algorithm (Multi-step GA) is applied for evaluating the gene candidates, with its effectiveness, simplicity and recursive evaluation abilities. A docking experiment using proposed system has been done in the sea successfully at Wakayama. Real-time recognizing performance is evaluated with observed data extracted from real-time video showing that the real-time recognition accuracy especially in the depth direction is increased by using dual-eye cameras and docking performance in sea has been confirmed.