Context-independent real-time event recognition: Application to key-image extraction

The significant increase of video data in various domains requires effective ways to represent video by its semantic content. This paper proposes a system to extract useful events defined by approximate but efficient world models, independently of context. Changes and the behavior of low-level features of the scene's objects are continually monitored. When certain conditions are met, events related to these conditions are detected. Proposed events are sufficiently broad to assist surveillance and retrieval of video shots. Extensive experimentations on more than 10 indoor and outdoor video shots containing a total of 6371 images including sequences with noise and coding artifacts have demonstrated the reliability and the real-time response (up to 10 frames per second) of the proposed system.