Motion monitoring and classification with center-biased motion estimation

A major problem in video surveillance is the robust detection of intrusive motion. A widespread technique is frame differencing, initiating an alarm procedure when the difference exceeds a predefined threshold. However, background motion and conflict motion are hard to distinguish. Full-search (FS) block-based motion estimation can be made more robust with respect to the rejection of very small motions, but it is still too sensitive and it has also the disadvantage of the high computational complexity. In our proposal we suggest to use a detection technique that is based on a new center-biased search technique and an original adaptive motion classification system. The proposed approach was validated via an implementation on a very long instruction word (VLIW) processor