A method for recognizing living activities in homes using positioning sensor and power meters

Abstract

To realize smart homes with sophisticated services including energy-saving context-aware appliance control in homes and elderly monitoring systems, automatic recognition of human activities in homes is essential. Several daily activity recognition methods have been proposed so far, but most of them still have issues to be solved such as high deployment cost due to many sensors and/or violation of users' feeling of privacy due to use of cameras. Moreover, many activity recognition methods using wearable sensors have been proposed, but they focus on simple human activities like walking, running, etc. and it is difficult to use these methods for recognition of various complex activities in homes. In this paper, we propose a machine learning based method for recognizing various daily activities in homes using only positioning sensors equipped by inhabitants and power meters attached to appliances. To efficiently collect training data for constructing a recognition model, we have developed a tool which visualizes a time series of sensor data and facilitates a user to put labels (activity types) to a specified time interval of the sensor data. We obtain training samples by dividing the extracted training data by a fixed time window and calculating for each sample position and power consumptions averaged over a time window as feature values. Then, the obtained samples are used to construct an activity recognition model by machine learning. Targeting six different activities (watching TV, taking a meal, cooking, reading a book, washing dishes, and other), we applied our proposed method to the sensor data collected in a smart home testbed. As a result, our method recognized 6 different activities with precision of about 85% and recall of about 82%.