3D egomotion from stereo cameras using constrained search window and bundle adjustment

We propose a method for egomotion estimation from images obtained by a pair of stereo cameras for the reconstruction of 3D scene. The proposed method improves feature matching by using two eliminating criteria to filter out unlikely matching points and restricts the search window according to the epipolar constraint. The method also improves calculated motion by using bundle adjustment to integrate all relevant parameters for optimization. Experiments have been performed to demonstrate the improvement of using the proposed egomotion estimation.