Collaborative data aggregation using inference tree for occupancy reasoning in visual sensor networks

Abstract

Video surveillance is a popular consumer application that is used for various purposes such as public safety, facilities surveillance, and traffic monitoring. In a general video surveillance system, video streams from cameras are sent to a control center and operators monitor the videos. But human operator monitoring of the views every moment of every day is almost impossible; so, smart surveillance systems are required, systems that are capable of automated scene analysis. There are a number of studies to enable smart video surveillance in a multi-camera network. Most of the studies, however, treat central processing approaches in which a scene analysis is processed inside a central server domain once all available information has been collected in the server. Such approaches require tremendous efforts in building the system and, moreover, limit the scalability. To accomplish scalable smart video surveillance, an inference framework in visual sensor networks is necessary, one in which autonomous scene analysis is performed via distributed and collaborative processing among camera nodes without necessity for a high performance server. In this paper, we propose a collaborative inference framework for visual sensor networks and an efficient occupancy reasoning algorithm that is essential in smart video surveillance based on the framework. We estimate the existence probabilities for every camera and combine them using the inference-tree architecture in a distributed and collaborative manner. We aim for practical smart video surveillance systems.