Model-based vision for automatic alarm interpretation

The further development of a previously reported knowledge-based vision recognition system for automating the interpretation of alarm events resulting from a perimeter intrusion detection system (PIDS) is reported. Measurements extracted over a sequence of digitized images are analyzed to identify the cause of the alarm. Models are maintained for alarm causes, and the scene and the measurements are matched with the models to derive an appropriate classification of the event. An overview of the initial classification system is provided, and modifications to the system are described. The results of applying the system to a range of real alarm events are presented, and further improvements are discussed.<>