Analyzing a real world scene sequence using fuzziness

A technique to separate images of non-stationary from stationary components in a TV-frame sequence of a real world scene had previously been reported [1]. Using an approach related to fuzzy set theory, this technique has been significantly improved. Analyzing TV-frames representing a downtown street intersection for one second, bitmasks corresponding to images of a car and a pedestrian as well as their trajectories have been extracted from this sequence - without using scene specific knowledge about the moving objects or the stationary scene components. Similar results have been obtained for other TV-frame sequences. This approach may contribute to an algorithmic solution for the "peripheral vision" problem, i.e. to detect a moving object in the field of view, and for the "attention focussing" problem, i.e. to decide which part of the images in a sequence to investigate in detail.