AI-Assisted Markerless Activity Tracking System for Supporting Aging and Wellness

This paper presents a mm-Wave radar-based system to create a novel method of tracking seniors and monitoring them using ambient wireless signals. Radar sensors, coupled with deep learning models, facilitate the identification of various physical activities without requiring physical contact or the use of wearable devices. The system employs range-Doppler maps derived from a real-life in-home activities dataset for training deep learning networks. The gated recurrent units (GRU) model is chosen for real-time implementation due to its optimal trade-off between speed and accuracy. The system achieves an overall accuracy of 93% for trained subjects, with 86% accuracy for classifying the in-home physical activities of a new subject. Additionally, the system captures the frequency of washroom use, sedentary, duration of sleep, active and out-of-home periods, the level of activity performed by the subject over a period of time, current activity state, and gait parameters.