Simultaneous Inference of Places, Activities, and Behavioral Classes in Maritime GPS Traces*

Previous work has shown that activities and places of interest can be extracted from GPS traces of human movements using behavioral models based on conditional random fields (CRFs) [3]. In this paper, we adapt and extend this work in two ways. First, we apply the framework to analysis of a vehicle-tracking maritime environment, analyzing GPS data from a 5 day surveillance of merchant marine ships conducting exercises in the English channel. Secondly, we expand the model to a perform a broader population analysis segmenting the population into several classes with distinct behavioral models. Empirical results show that our algorithm is successful in inferring locations of interest, but makes only coarse distinction in activity inference. In clustering behaviors, it successfully divides agents with highly localized activities from those servicing distant ports