Detection of activities of daily living with decision trees through a technical assistance system

Abstract

To cope with the increasing demands in care due to the aging society and the simultaneous lack of professional caregiversi a technical assistance system can help to monitor elderly people in their own homes and to support professional caregivers and caring relatives. The technical assistance system consists of a smart sensor with an omnidirectional camera on the ceiling of each room and additionally, other smart home sensors in an apartment for elderly persons. Based on smart sensor data, the positions, poses, and activities of the patient are detected with the help of machine learning (ML) techniques. In this work a temporal behaviour model of the patient is developed to recognize activities of daily living (ADL) such as eating, sleeping or emergency. For this, the actions (e.g., walking, sitting) and the location in the current room of the patient, as well as the data of other sensors in the apartment are used. This input data is fed into a trained decision tree of depth 14, which ultimately determines the patient's activity. The accuracy for detecting activities of daily living with the decision tree is 96.47%, where the activities can be detected in real-time.