本論文設計了一種應用於二軸馬達控制平台與輪型機器人移動平台的立體視覺里程計，可達到三軸旋轉與三軸位移的姿態估測效果。首先，本論文提出了一個估測立體視覺攝影機的內部參數方法，其可加強姿態估測資訊的準確性。此攝影機的內部參數是由立體視覺攝影機實際觀測到的資訊來估測，透過實際資訊得到的內部參數來做更佳符合攝影機的姿態估測。接著，透過估測到的攝影機內部參數，結合重新映射方法來設計一個有效的最佳姿態估測演算法，其使用立體視覺攝影機所觀測到的資訊來做連續影像間的累積姿態。本論文所提出的方法已經驗證在二軸馬達平台與輪型機器人姿態估測上，也驗證確實較一個現有演算法為佳。 This thesis proposes a stereo vision odometry algorithm used in two degree-of-freedom (DOF) pan-tilt and wheeled robot mobile platforms. The proposed algorithm achieves six-DOF camera pose estimation. To improve the accuracy of pose estimation, a simple and efficient calibration method is proposed to find the internal parameters of a stereo vision camera. Next, an optimal camera pose estimation algorithm is designed via a re-projection scheme based on the estimated camera internal parameters. Applying the proposed algorithm to a stereo vision camera system allows a platform equipped with the system actually estimating the pose change between two consecutive images, achieving the purpose of visual odometry. When applied on a two-DOF pan-tilt platform and a wheeled mobile robot, experimental results validate the performance of the proposed algorithm by comparing with an existing method.
