A knowledge-based approach to automatic alarm interpretation using computer vision, on image sequences

This paper describes the development of a knowledge based system which will be used to automate the interpretation of an alarm event resulting from a perimeter intrusion detection system. The knowledge-based system analyses a sequence of digital images captured before, during and after the alarm is generated. Additional data, pertaining to the alarm sensor, prevailing weather conditions and time-of-day are also available to assist the interpretation. In order to cope with the diverse nature of the different data sources, a knowledge-based approach is used to perform the interpretation. Models are maintained for a variety of possible alarm causes (human, animal, environmental, false etc.) and each model characterises a number of properties associated with that particular alarm source. The event data is interrogated by the KBS following the selection of a particular model.