Motion Estimation of a camera system (a.k.a. Ego-Motion estimation) has been an active field of research for decades. Today, often the term visual odometry is used. In vehicles, usually sensors for velocity and yaw rate (rotational speed around the height axis) are available.
The sometimes large rotational velocities around the other axes due to braking or bad roads (tilt and roll motion) are not measured. Such motions are interpreted as vertical motions of the tracked points as shown in the left sequence of the video.
6D-Vision helps to resolve the issue: from a few stationary 3D points one can determine the complete motion state of the vehicle (rotation and translation) precisely. Hernan Badino, a former PhD student of the team, has developed an run-time efficient method in 2005 to compute this motion. [Badino 06].
The video shows a result of this ego-motion estimation ob the right, where all occurring tilt and roll motions are compensated. Now all static points are determined as static, the few remaining motion vectors due to flow errors are removed by downstream algorithms.
Video: 6D-Vision without (left) und with (right) image-based ego-motion estimation