Event Recognition in Airborne Motion Imagery

A system is described that detects high-level events of interest in airborne motion imagery. The system starts by automatically extracting the tracks of movers from the imagery. It then uses tracks combined with context (cultural features, for example) to automatically detect low-level events. High-level events can be defined as patterns of lowlevel events. Exact or partial matches to these patterns of interest can be presented to an operator for further analysis.