A method for rotation matrix estimation in stereo visual odometry based on left and right rotation averaging

Abstract

For Stereo Visual Odometry (SVO), the rotation and translation of camera motion can be estimated simultaneously or separately in which rotation is extracted from an essential matrix. Most of state of art methods computing the parameters of rotation camera motion only use each pair of consecutive image frames on the left side. This leads to not leveraging the worth information of a pair of consecutive image frames on the right side. This paper presents an approach to leverage this information during computing rotation by averaging of value of rotations extracted from both left and right sides. The proposed method is evaluated on the KITTI dataset to verify the performance of this algorithm. The experimental result indicates that the proposed approach enhances about 10% accuracy compared to other methods in the same scenario.