仿生型自主式水下載具利用單眼視覺在已知環境中之導航研究

Abstract

本論文探討仿生機器魚使用單眼攝影機與電子羅盤及加速度計，在已知的水下環境下進行定位及導航。本文首先建立被觀測物體與單眼視覺的關係式，通過色彩空間演算法來對目標物做分類，再使用邊緣偵測找出目標物在影像中的位置與寬度，並利用這些資訊經由實驗估測出，機器魚視訊攝影機與已知目標物間的距離及角度關係。而估測出的相對關係即可做為機器魚的觀測資訊，此資訊整合方向及加速度的資料，與機器魚估計自身的運動關係即是機器魚之運動模型。此運動模型與觀測資訊可用來構成機器魚之延伸型卡曼濾波器定位演算法，以達成機器魚在已知的水下環境中進行定位之目的。最後，本文利用模擬水下機器人競賽的關卡當作已知環境來驗證此導航法之可行性。

This article describes a localization and navigation algorithm in a known underwater environment for a biomimetic robotic fish. The navigation algorithm combines information out of a monocular camera, an electronic compass, and accelerometers. The method of localization finds the relative position and orientation of a monocular camera with respect to an environment object by classifying the object in the image color space, then using edge detection to identify the position and width in the reference coordinate. Extensive tank experimental data were gathered to estimate the relative distance and angular relationship between the robotic fish and its environment object. Information of the relative position and orientation can be used as the observation data for the robotic fish by integration with acceleration data and the motion model of the robotic fish to calculate the motion estimation. An Extended Kalman filter localization algorithm was then formed for the robotic fish to perform self-localization in the known underwater environment. Finally, a simulated site for an international aqua robot competition event was used as an example to verify the feasibility of the proposed localization and navigation method.