Spatio-Temporal Reasoning Within a Traffic Surveillance System

Abstract

The majority of potential vision applications such as robotic guidance and visual surveillance involve the real time analysis and description of object behaviour from image sequences. In the VIEWS project we are developing advanced visual surveillance capabilities for situations where the scene structure, objects and much of the expected behaviour is known. This combines competences from image understanding, knowledge-based processing and real-time technology. In this paper we discuss the spatio-temporal reasoning which is of central importance to the system allowing behavioral feedback. In particular, we will elaborate the analysis of occlusion behaviour where we need knowledge of the camera geometry to invoke the occlusion region monitoring of vehicles plus knowledge of the scene geometry to maintain high-level models of possible trajectories for the occluded vehicles and to recognise the re-emerging vehicle(s).