Modeling of Expert Knowledge for Maritime Situation Assessment

In today's surveillance systems, there is a need for enhancing the situation awareness of an operator. Supporting the situation assessment process can be done by extending the system with a module for automatic interpretation of the observed environment. In this article, the information flow in intelligent surveillance systems is described and a detailed modeling of the situation assessment process is presented. The main contribution of this article is a probabilistic modeling of situations of interest. The result of this modeling is a Situational Dependency Network (SDN), which represents the dependencies between several situations of different abstraction levels. The focus is on a top-down approach, i.e., the modeling is done in a human-understandable way and can be done by maritime experts. As especially critical situations can change very fast in their characteristics and also they do not happen very often, the machine learning approach is not appropriate for detecting such situations, even if they are very powerful. Therefore, we present an approach, where expert knowledge can be included into a Dynamic Bayesian Network (DBN). In this article, we will show how a DBN can be generated automatically from the SDN. We mainly focus on the determination of the parameters of the model, as this is the crucial point. The resulting DBN can then be applied to vessel tracks and the probability of the modeled situations of interest can be inferred over time. Finally, we present an example in the maritime domain and show that the probabilistic model yields the expected results.