An Evidential Fusion Approach for Activity Recognition in Ambient Intelligence Environments

International audience With the growing emergence of ambient intelligence, ubiquitous computing, sensor networks and wireless networking technologies, "ubiquitous networked robotics" is becoming an active research domain of intelligent autonomous systems. It targets new innovative applications in which robotic systems will become part of these networks of artifacts to provide novel capabilities and various assistive services anywhere and anytime, such as healthcare and monitoring services for elderly in Ambient Assisted Living (AAL) environments. Situation recognition, in general, and activity recognition, in particular, provide an added value on the contextual information that can help the ubiquitous networked robot to autonomously provide the best service that meet the needs of the elderly. Dempster-Shafer theory of evidence and its derivatives are an efficient tool to handle uncertainty and incompleteness in smart homes and ubiquitous computing environments. However, their combination rules yield counter-intuitive results in high conflicting activities. In this paper, we propose a new approach to support conflict resolution in activity recognition in AAL environments. This approach is based on a new mapping for conflict evidential fusion to increase the efficiency and accuracy of activity recognition. It gives intuitive interpretation for combining multiple sources in all conflicting situations. The proposed approach, evaluated on a real world smart home dataset, achieves 78% of accuracy in activity recognition. The obtained results outperform those obtained with the existing combination rules.