A Robust Approach for Ego-Motion Estimation Using a Mobile Stereo Platform

We propose a robust approach to the problem of ego-motion estimation using a mobile stereo platform. Stereo is computed for every frame obtaining 3D points of the environment. In addition optical flow establishes correspondences between points in successive frames. A smoothness motion constraint is applied in order to detect flow vectors which are inconsistent with the current ego-motion. The optimal rotation and translation between the resulting clouds of static points is then computed using a closed form solution based on unit quaternions, providing the motion between the current and previous frame. Stabilization and a better estimation are achieved computing the observed motion between the current frame and many frames in the past in a multi-frame fashion. The integration of successive movements allows the reconstruction of the travelled path. Experimental results with sequences covering more than 1.5 kilometers of travelled distance are presented and compared with GPS and odometry.