The 6D-Vision principle to track 3D points from frame to frame and to determine position and speed is not limited to isolated points. It can be applied also to groups of image points, so called superpixels.
We use rectangular elements, so called Stixels as superpixels, to model the traffic scene compactly without relevant information loss, before we start the 6D-Vision process and subsequent applications. Thanks to a powerful optimization scheme  these Stixels adapt very well to the objects in the traffic scene. The image shows an example of the representation of an urban scene, the colors encode the distance (red=near, green=far). Subsequent steps such a s obstacle detection freespace computation or attention control don´t have to analyze half a million 3D points but only about 500-100 Stixels per image.
When tracking Stixels over multiple frames, the 6D-Vision principle allows to estimate their motion.
The video shows a situation at a traffic light, where the arows point to the expected position in 500ms, the color encodes the distance of the Stixel . Since the 6D-Vision filter step considers all image points in an integrative fashion, one obtains more accurate measurements of position and velocity than with isolated points - another great step foreward.
Current Research >