Characterization of Occlusion Situations Occurring in Real{world Traac Scenes

During evaluation of real{world traac scenes, we often encounter the situation that the vehicles under scrutiny are temporarily occluded by dynamic or stationary scene components. Vision{based tracking algorithms often fail in tracking vehicles under such conditions. Contextual knowledge about occlusions is expected to facilitate the vehicle tracking process. We collected a set of potential occlusion situations, described by primitive predicates which are modeled by means of fuzzy sets, and set up a transition diagram in order to obtain all possible transitions between these occlusion situations. We are thus able to characterize the temporal development of occlusions. In this contribution we compute such occlusion characterizations from real{world image sequences as an intermediate step towards algorithmic formulations of natural language descriptions of activities captured by image sequences.