A Bayesian approach to the detection of hazardous shipping activity

Tracking of ships by means of various heterogeneous sensors is well understood and regularly practiced. But in order to carry out surveillance for security purposes, the tracking of single units is only the fundamental step for the detection and classification of activities. The decision about potentially hazardous activities shall not only be based on single tracks but rather on the relation of tracks from multiple objects. Particularly along the coastal strip and in the vicinity of ports the mostly crowded ship movement poses a huge challenge to activity detection and classification. In this article we describe the information flow in an intelligent surveillance system and clarify the separation of the real world and the world model, which is used for the representation of the real world in the system. The focus of this article is on modeling situations of interest in surveillance applications and inferring them from sensor observations. For the representation in the system, concepts of objects, scenes, relations, and situations are introduced. Situations are modeled as nodes in a dynamic Bayesian network, in which the evidences are based on the content of the world model. Several methods for inferring situations of interest are suggested, which make use of the underlying network modeling. Due to this modeling, we get a probability of all the situations in the network in every time-step. By collecting more evidences over time, the probability of a specific situation is either increasing or decreasing. Finally, we give an example of a situation of interest in the maritime domain and show how the probability of the situation of interest evolves over time.