Using Concurrent Hidden Markov Models to Analyse Human Behaviours in a Smart Home Environment

This paper addresses learning and recognition of human behaviour models from multi-modal observations in a smart home environment. The proposed approach successfully implements concurrent Hidden Markov Models that identify the occurring situation. This approach corresponds to the highlevel part from a framework to obtain high-level classification of human behaviour analysis. The results were obtained for a smart home environment, where cameras, microphones and a PMD sensor were deployed. The sensory information was first fed to the low-level classification stage, where it was analysed by four different classifiers which generate the observations to the high-level classification stage. For each situation an HMM is used, allowing the fusion of the data provided by the different sources present in the low-level classification stage. This approach proved to be highly scalable, since the recognition of new situations can be accomplished by means of adding the adequate HMM.