Motion Analysis In Compressed Video - An Hybrid Approach

In this paper, we propose a technique for the automatic extraction of moving objects from compressed domain video. We propose the use of a spatio-temporal filter for filtering the motion vectors and a hybrid approach to exploit both compressed domain processing and spatial domain processing to meet the tradeoff between computation and accuracy. The experimental results shows that our approach can detect objects with size smaller than one macroblock with accurate shape of the object.