[[alternative]]Path Planning and Obstacle Avoidance Based on Stereo Vision for Autonomous Guided Cart

碩士電機工程學系[[abstract]]本文研究具有機械視覺在一般室內環境中能自我導航的無人駕駛車，研究的內容包括雙眼主動攝影機控制、3D場景重建、障礙物偵測及路徑規劃。文中提出雙軸平行運動機構的設計及控制方法，將兩部CCD攝影機架設在此機構上，能精準的進行上下仰俯及左右?轉的取像控制。藉由兩部攝影機取像，利用視差的原理建立運動環境的3D場景，可以獲得運動空間的完整資訊，定位出障礙物相關的位置，並尋找出可行的運動路徑。以Hough Transform為基礎設計可以取得走廊左右邊線及走廊盡頭終止線的處理方法，用以界定出可行動空間及障礙物。另外，使用可變的樣板建立一套標誌辨識系統，辨識目的地的標誌，做為無人駕駛車的導航終點。依據重建的3D場景可找出與環境模型相對應的起始點、障礙物和目的地的導航地圖。辨識結果配合無人駕駛車運動學來規畫運動路徑及避障路徑。得到運動路徑及避障路徑後，路徑追蹤控制器及視覺導航控制器控制無人駕駛車依車速命令進行導航控制到達目的地。最後，提出實現無人駕駛車的軟硬體系統，包括車體結構、馬達控制器、運動控制及取像機構伺服馬達控制等。並藉由實驗來驗證視覺導航的可行性。[[abstract]]This thesis investigated an autonomous guided cart (AGC) with machine vision. This mechanism has self-navigating ability in indoor environments. This research includes binocular camera control, 3D scene reconstruction, obstacle detection and path planning. The two axis parallel manipulators and the motion control are design for precisely manipulating the binocular cameras with pan and tilt grabbing ability. The disparities in the binocular images are used to render the 3D navigation scenes in which the obstacles are detected and the available movement paths are found. Base on the Hough Transform image processing approach, the left, right and end of a corridor are identified and the movable space and obstacles detected. The variable template method was used to design a pattern identification algorithm for the AGC to recognize the destination pattern. The navigating map in which the correspondence paths of departures, obstacles and destinations are obtained using 3D navigation environment rendered models. The results obtained using the navigation map and the cart kinematics are employed to plan the movement and avoidance paths for the AGC. After obtaining the movement and avoidance paths, the trajectory tracking and vision servo controller uses the pre-defined speed command to control AGC to reach the destination. The vision servo controlled AGC hardware design and software implementation, which includes the cart body structure, motor drivers, motion control and two parallel axis servo motor control manipulators are proposed. The experimental results demonstrate that vision navigation control is feasible