Event detection using multimodal feature analysis

This paper presents an event detection framework using multimodal feature analysis. In this framework, multimodal features are extracted from video data and then analyzed to generate various mid-level concepts, such as video shot, face appearance and so on. Two schemes, the logistic regression and Bayesian belief network, are then employed to fuse the information obtained from multimodal feature analysis and detect the video events of interest. We aim to use this framework as a general template for event detection in different video domains. Experimental results on various test videos in different video domains suggest that the proposed event detection framework is promising.