Motion pattern based video classification using support vector machines

Semantic classification is an effective approach to the management of vast digital video data. We propose a new semantic classification scheme based on motion patterns. With such a scheme, the motion patterns in video clips can be effectively mapped to semantic conceptions. Motion texture (see Ma, Y.F. and Zhang, H.J., "Motion Texture: A New Representation for Video Content", Technical Report, Microsoft Research, 2001) is employed as motion pattern descriptor, which can be extracted from shots or video clips. By using kernel support vector machines (SVMs), we have devised an optimized multi-class classifier to link low level features with conceptions. Experimental results indicate that our approach is an effective solution for motion pattern based semantic video classification.