An integrated stereo ego-motion estimation framework with sparse depth map

In this paper, we propose a novel method to accurately estimate the arbitrary motion of a calibrated stereo rig from a noisy sequence. In the proposed method, a projective camera model is used which is appropriate for scenes where the objects are close to the camera or where there is depth variation in the scene. We propose a feature-based method that estimates large 3D translation and rotation motion of a moving rig. The translational velocity and acceleration and angular velocity of the rig are then estimated using a recursive method. In addition, we account for different motion types such as pure rotation and pure translation in different directions. In our studies, we assume that the rig motion is noisy, i.e., the acceleration and velocity of the camera are not perfectly constant. Our experimental results show that we obtain accurate estimates of rotation matrix and translation vector parameters across different test-cases with large and small baselines. For long sequences, the estimated motion parameters are within ±10 pixels.