学位論文の要旨
Abstract of Thesis 研究科
School
自然科学研究科
専 攻
Division
産業創成工学専攻
学生番号
Student No.
51430710
氏 名
Name
許 弘毅
学位論文題目 Title of Thesis（学位論文題目が英語の場合は和訳を付記）
Stereo-vision-based AUV Docking System Adapted to Real-sea Environment
（実海域環境に適応したステレオビジョンAUVドッキングシステム）
学位論文の要旨 Abstract of Thesis
Autonomous underwater vehicles (AUVs) play an important role in human society in different applications such as inspections of underwater structures (dams, bridges), ship hull inspections, and scientific studies of a deep ocean. However, underwater vehicle operations are limited to activities that can be completed in short duration time, which is caused by the limited power capacity of batteries set on the vehicles. Even though advanced technology related to power devices provides longer operation time, underwater vehicles must come back to a surface vessel for recharging when operations take more than a couple of days. A recharging function that needs the AUVs to be docked to the recharging station at sea bottom can enable the AUVs to operate in extended period in an actual sea without returning surface vehicle for recharging. Therefore, underwater recharging system with a docking function have been researched in various approaches. Most research have used single camera for docking. On the other hand, the precision of distance measurement in the camera depth direction is not enough to accomplish the docking by AUVs in which high homing accuracy is indispensable.
To overcome this problem, the author has proposed a new 3D-perception-based move-on-sensing (3D-MoS) system with stereo vision cameras for underwater docking operations. The visual servoing of an underwater vehicle using the parallactic nature of stereoscopic vision and 3D-model-based recognition method and real-time multi-step genetic algorithm (RM-GA) was initiated by our research group for the 3D-MoS.
This thesis proposes the improved robust recognition system of real-time 3D pose tracking against the lighting condition changes. The fitness function of the RM-GA has been modified by adding a heuristic rule that increases the evaluation value when all color balls of the model overlap real 3D marker in the camera images. This approach enables our docking system to apply to the natural lighting environment that changes every moment during actual docking experiments. After modifying the recognition method, the effectiveness of the proposed fitness function adaptive to the changing lighting environment has been confirmed in the outdoor pool environment.
Even though, the abilities of the proposed system have been limited to some extent because of fluctuations of strength and changings of direction of the sea currents. According to the authors’
knowledge, no existing study has conducted the docking using stereo-vision-based real-time visual
Name 許 弘毅
servoing with tolerance performance against external force caused by current changes and illumination varieties. The docking system in real-sea environment cannot avoid the influences of the external force by an ocean current and the natural lighting environment that changes every moment.
To overcome the above situation, a current-adaptive docking station was devised. The current- adaptive docking station can change its docking direction following to the current direction, through which the burden of controlling the underwater vehicle's heading can be reduced. Next, the effectiveness of the docking station adaptive to the changing current direction has been proven by successful repeated docking experiments in the environment with fluctuating current and changing lighting condition with turbidity disturbances in real sea.
In the last part of the thesis, the inevitable problem of AUV operation, that is, resetting the accumulated position errors is solved by the proposed system. While returning to the docking station for recharging battery, AUV utilizes the navigation system, such as the Inertial Navigation System (INS) to approach the docking station. However, the success rate of the docking operation of AUV decreases as the INS accumulated error increases. To solve this problem, the proposed docking system is combined with the INS. The accumulated error can be eliminated by utilizing the docking information, that is, current position and orientation with accumulated errors are replaced by correct and predefined position and orientation of the docking station. Experimental result shows that the proposed system is capable of docking the AUV and resetting the integrated INS error. This demonstrates the effectiveness and functionality of the proposed system applied to the AUV battery recharging operation, thereby providing a reliable solution to extend the operation time of the AUV.
