To survey large areas or buildings, it is necessary to install an extensive amount of sensors. In conventional surveillance systems, all signals from various sensors, such as cameras, microphones, RFID detectors and light barriers, have to be evaluated manually by human operators, which is cost-intensive and error-prone. Present approaches for a more automated surveillance mainly focus on the automated exploitation of a single sensor signal, enabling the system to notify the operator of primitive events. The most critical threats however, such as terrorism and industrial espionage, can only be detected and avoided, if information from multiple sensor signals is fused together. Therefore, an approach to represent the relevant information extracted from sensor signals, fused into a single comprehensive, dynamic model of the monitored area is presented. The proposed object-oriented world model (OOWM) is part of the semi-autonomous surveillance system NEST, developed at Fraunhofer IITB. The main focus of this research project is to migrate from a sensor-centered view to a task- and objectcentered view, with the aim to focus the system on the applicationrelevant information.
The OOWM is preconfigured with static prior knowledge, such as the spatial data of the area or building to be supervised. Dynamic information is collected by arbitrary signal exploitation algorithms (e. g.person tracking) and fused into a consistent representation inside the world model by means of Bayesian fusion. As an obvious benefit, the resulting object-oriented representation of the dynamic situation provides the operator with a tool to easily overview the situation, detect critical situations early and initiate appropriate countermeasures. Furthermore, the world model is a key component to enable an easy and seamless integration of new sensors and to develop high-level algorithms that are able to execute surveillance tasks autonomously.