An image processing board with an MPEG processor and additional confidence calculation for fast and robust optic flow generation in real environments

This paper describes a vision system based on a PC-board which can calculate a sparse but robust optic flow at frame rate. A correlation chip, which was originally designed for MPEG video compression, is used to calculate displacement vectors between blocks of pixels in consecutive frames. The main disadvantage of similar approaches, the noisiness of the displacement field in areas with weak structure, is compensated by a computational inexpensive confidence criterion, which is calculated for each vector in hardware. The performance of the criterion and the improvements in the flow field are demonstrated by experiments.