6D-Vision‎ > ‎

The Principle

The image sequence on the right shows a collision risk scenario, since the kid will be in 1 second at the same position as the vehicle where the cameras are mounted. In order to detect this, the Computer has to determine both the position and speed of the kid. 6D-Vision uses stereo vision to determine the 3D position of image points. In addition an image point is tracked over several images in a sequence to measure speed (Optical Flow). Since we obtain a 3D position at  this point again, we can determine a 3D-velocity with suitable filtering. Details of this filter are explained in [3] and [5]. 3D position and 3D motion together yield the six dimensions of the method.

If the vehicle is stationary, the task of determining other object´s velocity is solved. If it moves, the system determines the relative motion between object and moving camera system. This means, that all stationary objects virtually move with the velocity of the observing vehicle. As long as stationary objects are visible in the scene, one can compute the ego-motion based on 6D-Vision info and determine the absolute object motion over ground.

The video below shows a result of 6D-Vision. The color of the analyzed image points encodes the distance with red being near and green being far. The arrows visualize the velocity of the points and point to the expected position in 0.5s  On modern computer hardware, 10000 or even more image points can be processed. A further development of 6D-Vision  to estimate position and motion for every image point is described in Section Dense6D.