Complex event recognition with uncertainty reasoning

The goal of complex event recognition considered in this paper is the automatic detection of complex high-level events in videos. This is a difficult task, especially when videos are captured under unconstrained conditions, with poor lighting, heavy background clutter and occlusion. In this paper, we propose a hierarchical knowledge-based framework for complex event recognition. The video event knowledge represents an arbitrary complex spatio-temporal event as a hierarchical composition of simpler events in a natural way. Uncertainty reasoning procedures are applied to interpret low level event descriptions according to the video knowledge base in order to recognize high level scenarios.