High accuracy visual odometry using multi-camera systems

This work proposed an improvement for the visual odometry based on a new hardware design concept. Generally, a visual odometry system relies on one stereo camera to detect the 3D features which are used as input for the ego motion estimation process. However, such design usually faced with the motion ambiguity problem resulting in poor estimation of the ego motion. The proposed multi-camera system solves this problem using an intuitive hardware design that enables a simple yet effective way to eliminate the motion ambiguity problems. Experiment results show that the proposed system yields better ego motion estimation compared to the conventional single camera system.