Efficient Measuring of Various Motions in MPEG Videos

There is a need to analyze motions in video sequences to characterize them since motion is a representative feature that distinguishes videos from other multimedia. However, the existing approaches can only provide very preliminary classification of motions. To support in-depth measurement of various motions, we need to distinguish the different motions generated by the different sources. In this paper, we explicitly distinguish camera and object motions in a shot. In addition to distinguishing motions themselves, we propose new techniques to measure not only amount but also direction of each motion distinguished, and express them into comparable forms. Our preliminary experimental studies indicate that the proposed techniques are promising.