Perceiving context in an immersive environment: a multisensor approach Stefano Piva, Luca Marchesotti, Carlo S. Regazzoni D.I.B.E., University of Genoa, via Opera Pia 11a, 16145 Genova, Italy carlo@dibe.unige.it Abstract. The proposed paper is mainly focused on a novel architecture for immersive multisensor data fusion. The use of agents and complex adaptive classification methods such as Self Organizing Maps (SOM), allows one to provide the system with the sufficient capability to organize information into high level abstraction contents. This knowledge or awareness of the environment can be exploited by the system itself in order to react and have an influence on the observed domain. The focus of the research carried out is concentrated on the context information defined as a set of features related to the environment, which are not created explicitly to be input to the system. 1 Introduction Ambient Intelligence is assuming more and more complex meaning. In recent years, due to the improved power and ability of the digital systems, this idea has become a concrete reality in research world. Today's trends in AmI oriented Machine Vision techniques go in the direction of systems capable not only to observe and analyze data coming from multiple and heterogeneous sensors, but also somehow to react to appropriate stimula in order to show a certain degree of “awareness“, “intelligence“ (context-dependent adaptavity) and “natural interaction” to the user [6]. In the vision proposed in this paper, AmI systems are defined as a set of virtual entities characterized by three fundamental capabilities: analysis, awareness, interaction. ISTAG gives in [1] a more formal definition of Ambient Intelligence (AmI) that points out as it should provide technologies to support human interactions and to surround users with intelligent sensors and interfaces. According to this, Brooks in [2] states that an Intelligent Environment has to make computation ``ready- at-hand'', putting the computers out into the real world of people more than people into the virtual word. In [3] Starner focuses a key issue of Wearable computing Systems which plays a fundamental role also in Ambient Intelligence: the capability of context sensing. Rhodes in [4] states that given a user and a set of goals, context is represented by a set of features related to the environment which are not created explicitly to be input to the system. According to [5] at least four different areas are involved in the design of AmI architectures: person identification, surveillance/monitoring, 3D methods, and smart rooms/perceptual user interfaces.