Visual Odometry using Ground Plane in Dynamic Untextured Environment

In this paper, we propose 6-Degrees-Of-Freedom visual odometry system which fused feature based visual odometry and the normal vector information of the ground plane. Unscented Kalman Filter is used for the sensor fusion to accomplish robust and accurate localization. The error of roll and pitch angle are reduced by using not only the 3D features but the information on the field of ground plane. As a result, our VO system enabled robust and precise 6DOF localization also in untextured dynamic environments. In feature-based visual odometry, robustness and accuracy are improved by using RANSAC, three-point algorithm, and key frame adjustment. In order to extract ground plane, the stereo homography matrix is used. To calculate the homography matrix robustly, we use RANSAC. We present and evaluate experimental results for our system in dyamic outdoor environments, and the effectiveness of our technique is shown.