Representation and Recognition of Events in Surveillance Video Using Petri Nets

Detection of events is an essential task in surveillance applications. This task requires finding a general event representation method and developing efficient recognition algorithms dealing with this representation. In this paper, we describe an interactive system for querying surveillance video about events. The queries may not be known in advance and have to be composed from primitive events and previously defined queries. We propose using Petri nets as both representation and recognition methods. The Petri net representation for users´ queries is derived automatically from simpler event nets. Recognition is then performed by tokens moving through the Petri nets.