自主式水下無人載具在視覺智能導引系統目標追踪的實現

REALIZATION OF OBJECT TRACKING FOR A VISUAL INTELLIGENT GUIDANCE SYSTEM OF AUTONOMOUS UNDERWATER VEHICLE

本研究旨在以人工智能為基礎建構一套自主式水下無人載具的視覺智能導引系統，並應用在水下目標物體識別與追蹤。藉由在視覺智能導引系統導入Mask R-CNN深度學習演算法及判別周遭即時特徵訊息，自主式水下無人載具將不再受限於無法依環境變動影響之特徵資料（邊緣、色彩、強度與面積等），更適合在水下環境中執行目標物識別與追蹤。視覺智能導引系統可即時自載具艏艙得到連續影像資訊，進行自主式水下無人載具周遭環境視覺資訊之演算與識別。另外，藉由多種空間座標定義轉換過程，演算自主式水下無人載具與目標物體之相對距離及方向角，再經由模糊控制方法計算航速與航向之操控係數，構成視覺智能導引運動操控架構。為驗證視覺智能導引系統之功能，本研究於國立成功大學系統及船舶機電工程學系穩定性能水槽及拖航水槽進行一系列試驗，並在0.5 m/s的拖航目標物之移動速度下，視覺智能導引系統之目標追踪影像橫向與縱向座標偏差平均值可達9 Pixel及24 Pixel，控制在視覺空間2.81%、10%範圍內，成功地驗證了視覺識別與導航的自主動態控制，完成自主式水下無人載具目標識別與追蹤能力之實現。

This study aims to develop a visual intelligent guidance system (VIGS) based on artificial intelligence for object recognition and tracking of autonomous underwater vehicle (AUV). By introducing the Mask R-CNN deep-learning algorithm into the VIGS and identifying the surrounding real-time feature information, the AUV will no longer be limited by the feature data subjected to environmental changes, so it is more suitable for object recognition and tracking in underwater environments. The VIGS can obtain continuous image information from the AUV's bow cabin in real time, and perform the calculation and identification of the visual information in the surrounding environment. In addition, the transformation process is defined by a variety of spatial coordinates, which can be used to calculate the relative distance and heading angle between the AUV and the target object. The control coefficients of sailing speed and heading angle are defined by the fuzzy logic controller, which can constitute the motion control architecture in the VIGS. In order to verify the performance of the VIGS, this study will conduct a series of experiments in the stability tank and the towing tank at Department of Systems and Naval Mechatronic Engineering, National Cheng Kung University. Meanwhile, an object will be designed as an underwater target to realize the object recognition and tracking capabilities of AUV.