高速轉彎自主式水下載具動力模型鑑定與 單眼視覺導航研究

Dynamic Modeling and Monocular Image-Based Pose Tracking for an AUV in Power Turn

邱奕倫(YI-LUN CHIU)

指導教授 ： 郭振華

國立臺灣大學/工學院/工程科學及海洋工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201600174

摘要

本論文探討兩項水下機器人基本問題：動力模型的建立以及載具位置的追蹤定位。本載具使用轉軸推進器，可於高速下執行橫向轉彎，並具備胸鰭進行急煞與深度控制動作。為了能夠預先估測載具動作，我們基於拉格朗日原理推導本載具動力模型。本文研究利用上方攝影系統來追蹤紀錄載具船體上的位置標記。進而取得載具的縱移、橫移與轉角速度等等數據，並應用非線性最佳化方法估測係數。經由比較模擬與實驗可知此動力模型具有相當的準確度與可靠性。
在載具的追蹤定位方面，我們採用單眼視覺方法來追蹤載具的位置與方向。我們使用前視鏡頭進行觀測，並提出一個新穎的即時最佳化估算方法。在已知地圖的環境中，使用粒子濾波器方法估測載具位置。特別的是將擴增實境的技術應用於量測模型中，這個量測方法為濾波器提供重要因子的計算方式。我們的方法已經過長時間水下巡航的驗證，實驗顯示該方法為穩定且高效能，即時提供水下載具位置與姿態。

關鍵字

自主式水下載具 ； 水下導航 ； 動力模型 ； 單眼視覺 ； 序列式蒙特卡羅定位演算法

並列摘要

This work investigates a development of a highly maneuverable AUV that has a high maneuverability to perform power turns. Two fundamental problems are addressed in this paper, which are the dynamic modelling of this AUV and pose tracking method by vision system. The vehicle has a rotatable stern propeller for horizontal turning at high speed, two paddles for the braking and ascending/descending. A motion model is firstly derived to predict the motion of the body. The dynamic equations are derived based on the Lagrange principle. Added mass coefficients are estimated using the equivalent ellipsoid method. A tank environment with an overhead camera system is utilized to record marker positions on the vehicle body. The iterative Lucas-Kanade method is applied for the tracking of the AUV.
To track the vehicle’s position and orientation for autonomous navigation, we introduce a monocular image-based approach. Our approach is developed for an underwater environment which with fewer features and low visibility. We present a novel real-time optimizing estimation method which bases on the forward-looking camera for observing. The sequence Monte-Carlo method is used for estimating the pose of body. In particular, the augmented reality technique is involved to the measuring process, this measuring method provide the reliable estimation for importance factor. Our approach was verified by long time cruise in a water tank. Experiment data indicates that it is robust and efficient for the real-time position tracking of the robot.

並列關鍵字

autonomous underwater vehicle ； dynamic modeling ； monocular vision ； Sequential Monte Carlo algorithm ； augmented reality

參考文獻

[9] 郭振華，邱柏昇, "仿生型自主式水下載具利用雙魚眼攝影機在已知環境中之導航," 國立臺灣大學工程科學及海洋工程研究所碩士論文, 2012.

[23] 郭振華，王偉翰, "使用側掃聲納掃描線輔助無人水下載具建立導航地圖," 國立臺灣大學工程科學及海洋工程研究所碩士論文, 2009.

[2] C. Connolly and T. Fliess, “A study of efficiency and accuracy in the transformation from RGB to CIELAB color space,” IEEE Transactions on Image Processing, vol. 6, pp. 1046-1048, Jul 1997.

[4] V. Kopman, J.Laut, F. Acquaviva, A. Rizzo, M. Porfiri, “Dynamic modeling of a robotic fish propelled by a compliant tail,” Oceanic Engineering, IEEE Journal of, 40(1), 209-221, 2015.

[5] T. I. Fossen, Nonlinear modelling and control of underwater vehicles, 1991.