画像の分散値に基づく移動ロボットの位置推定手法選択法の検討

抄録

Visual Odometry(VO) is a technique to estimate the position of a robot by using images. Compared to the wheel odometry, the VO is more accurate because it does not depend on the wheel slip. In a smooth surface environment such as sandy terrain, however, the VO fails to process because of the lack of feature points. In such an environment, it is better not to execute VO in order to reduce the processing load. Therefore, this paper proposes a method that decides whether VO is executable or not based on variance of image. The Random Forest-based classifier is trained using the images and archives high accuracy.