Improving recognition accuracy for activities of daily living by adding time and area related features
Yutaka Arakawa, Keiichi Yasumoto, Krita Pattamasiriwat, Teruhiro Mizumoto · 2017
Recognizing the activities of daily living (ADL) of residents in housing is indispensable for operating Daily Life Support Services such as Elderly Monitoring, Smart Home Automation, and Health Support. However, the existing methods have various problems: invasion of privacy, limited target activities, low recognition accuracy, initial installation cost, and long recognition time. As our prior work, we proposed a real-time ADL recognition method using indoor positioning sensor and power meters. We got a result that the method can recognize ten types of ADL with the average accuracy of 79%. However, the accuracy of some activities such as work/study and bathroom-related were not satisfactory. In this work, we aim to improve the accuracy of our prior method by newly adding several new time and are related features such as time slot when activity occurs, staying time in the same area, and previous position. As a result, we could achieve 82% of average recognition accuracy for 10 different activities.