Extraction of motion information is an essential step in representation and classification of the contents of video which require derivation of both spatial and temporal characteristics of objects appearing in the scene. Based on a low cost extension of the video compression algorithm MPEG, specifically its motion compensation component, we present a mechanism for extraction of movements of objects. Object motion, along with other semantic information, will lead to the generation of computer understandable representation of video contents. The talk will include a demo of a video analysis and retrieval system.