Using low-level sensory data to recognize events in a smart home

Abstract

The recent growth of smart, low-cost sensor devices combined with the ubiquity of WiFi and cellular networks have given rise to new opportunities for smart home applications in support of independent living of elderly people. Homes that are fitted with smart devices provide new conveniences, comfort, and safety for its inhabitants. To provide the needed support for independent living, the smart devices must be able to make predictions about the activities in the home and the residents’ needs and they must also be able to easily integrate into a home without costly alterations. This paper discusses the implementation and evaluation of a new sensor device that collects unobtrusively acoustic and other low-level sensory data and applies machine learning techniques to recognize events in the home. The device integrates into a prototype smart home system built in our lab to perform activity recognition in an effort to assist and improve the well-being of residents in a home. Results of our new sensor device suggest that events in the home can be recognized with high accuracy but that improving the accuracy for some events requires additional data.