Using a Hidden Markov Model for resident identification

In smart home environments, it is highly desirable to know who is performing what actions. This knowledge allows the system to accurately build individuals' histories and to take personalized action based on the current resident.Without a good handle on identity, multi-resident smart homes are less effective when used for medical and assistive applications. Most smart home systems either have a single occupancy requirement, or rely on a wireless or video device to identify individuals. These requirements are too burdensome in some situations, which can limit the deployment of smart home technologies in environments that would derive benefits from them. This research work introduces the use of passive sensors and a Hidden Markov Model as a means to identify individuals. The result is a passive, low profile means to attribute individual events to unique residents. For this work, two different pairs of individuals living in a smart home testbed are used to evaluate the tools. The data used is from unscripted, full time occupancy and annotated by the residents themselves for accuracy. Lastly, the Hidden Markov Model approach is compared and contrasted against a prior Naive Bayes solution on the same data sets.