Learning individual roles from video in a smart home

This paper addresses learning and recognition of individual roles from video data in a smart home environment. The proposed approach is part of a framework for acquiring a high-level contextual model for human behaviour in an intelligent environment. The proposed methods for role learning and recognition are based on Bayesian models. The input is the targets and their properties generated and tracked by a robust video tracking system in the environment. The output is the roles "walking", "standing", "sitting", "interacting with table", "sleeping" for each target. A Bayesian classifier produced good results for a framewise classification of these roles, while a hidden Markov model had even better performance taking into account a priori probabilities of roles and role transitions. A support vector machine produced best classification results. The classifiers had, however, problems to distinguish ambiguous roles like "walking" and "standing" in the environment. The obtained results permit to pass to the next step in future work: learning and recognizing relations and situation