6D-Vision‎ > ‎

Optical Flow

For 6D-Vision the perception of motion is crucial. One needs to compute accurate numbers for the so called "Optical Flow" - the temporal displacements of image points in an image sequence. The typical optical flow methods known from literature determine only small displacements. However, especially in areas close to the vehicle or in curves, large displacements occur.  The algorithm developed by Daimler in 2004 solves the problem for the first time [9]. Arbitrary large displacements are computed with a constant, small computation time and reliable results are obtained even when driving fast or under adverse weather conditions. The image on the right depicts the computed displacements where the color encodes the magnitude of the displacement - the  warmer the color the larger the displacement.

Recently, the Daimler researchers developed a new algorithm [15] that computes the displacement for every pixel, even for large displacements as shown above. The algorithm is implemented on a modern graphics card (GPU) and exploits the stereo vision information. This enables the computation of 6D-Vision at every image point (Dense6D) and will boost the performance of 6D-Vision further.