Passive wireless sensing for unsupervised human activity recognition in healthcare

Physical activity classification is an important tool for various applications such as activity of daily living (ADL) recognition and fall detection. Additionally, the non-contact nature of radar systems provides minimally invasive sensing platform. Doppler-based radar has been used for activity classification in the past. However, most of these studies considered supervised classification which requires labeled training data sets. In this paper, we propose a novel procedure of using micro Doppler radar for unsupervised classification with Hidden Markov Models (HMM). A low-complexity time alignment method for capturing activity is developed and an Elbow test has been adopted for model selection. Test results confirm the efficacy of the selected feature set and the proposed methodology. The results prove the proposed system can deliver a very good performance in ADL recognition tasks.