本論文以視覺里程計(Visual Odometry,VO)架設於二輪式移動機器人。首先以雙眼鏡頭作為感測器於未知環境中擷取影像，再於影像中使用加速強健特徵(speeded up robust features, SURF)搜尋影像上之地標。地標透過立體視覺求算三維座標，並用三點透視(perspective-3-point, P3P)問題與隨機取樣一致(random sample consensus, RANSAC)演算法求算出攝影機位置，達到視覺里程計的目的。最後實驗用機器人做簡單路徑行走，驗證此視覺式里程計系統為可行。 In this paper, a visual odometry is developed on the two-wheeled mobile robot. First, the binocular stereo camera is regarded as the sensor for the visual odometry to capture images in unknown environment. Then the landmarks of interest in images are detected by speeded-up robust features (SURF) algorithm. The location of the landmark in Cartesian coordinate can be derived from stereo geometry. The position of the camera can be determined by perspective-3-point (P3P) method and random sample consensus (RANSAC) algorithm. Therefore, the visual odometry is achieved. Finally, simple path navigation experiments demonstrate the effectiveness of the proposed visual odometry system.