Event Detection and Analysis from Video Streams

We present a system for event detection and analysis from video streams. Our approach is based on a detection and tracking module which extracts moving objects trajectories from a video stream. These trajectories, together with a rough description of the scene, are then used by the behavior inference module in order to recognize and classify object motion. The hierarchical tasks are performed on a buffered set of frames in order to provide accurate results by taking into account the temporal coherence of moving objects