摘要
本文提出一個用單一攝影機建立出週遭特徵並利用特徵找到攝影機位置後控制自動導航車的方法，在文中利用SIFT(Scale Invariant Feature Transform)找尋出週遭特徵點建立環境資料庫，且利用RANSAC(Random Sample Consensus)找出仿射轉換矩陣(affine transform matrix)得到即時影像和資料庫之相對位置。本文設計一個根據仿射轉換矩陣而調整自己的方向和速度的控制方法，到達關鍵點後做指定之任務，可以在室內或室外成功航行辨識週遭環境。中途若有障礙物時會進入短程VSLAM(Visual Simultaneous Localization and Mapping)，且使用本文提出短程VSLAM特徵點匹配的改良方法，可加速且精確找出畫面中實際特徵點位置，進一步利用預測和實際特徵位置可以調整擴展卡曼濾波器參數，推算出較精確特徵點座標，並且推算車體座標，在一段時間後可以有效建立出障礙物特徵地圖和自我定位，且在進入目標距離後進行避障。
並列摘要
The thesis proposes a method for a robot which uses map of the environment to localize itself accurately. A vision-based mobile robot localization and mapping algorithm which use SIFT (scale invariant of feature transform) to select suitable visual landmarks proposed in the thesis for mobile robot localization and database building, and RANSAC (Random Sample Consensus) is used to find affine transformation matrix and robot localization in image sequences and visual database. The control method of speed and direction of robot is developed to affine transformation matrix, and do designated tasks at specified destination. A robot can successful navigates and localizes in outdoor or indoor by using this method. It enter VSLAM (Visual Simultaneous Localization and Mapping) which will begin to find effective obstacle features when detect obstacles in navigation. The thesis proposes an improved method of VSLAM feature match in short distance which can be accelerated and accurate to find actual feature position in the image. Predicted feature's position and actual feature's position can updata EKF(Extended Kalman Filter) model to calculate accurate feature position in next time. VSLAM effectively build feature map and mobile robot localization at a period of time, and avoid the obstacle after entering the target distance.