Automated Fall Detection in Smart Homes Using Multiple Radars and Machine Learning Classifiers

Falls pose a significant risk, especially among elderly persons. Recently, radar sensors have been explored for fall detection. In this study, an attempt has been made to classify fall detection using multiple radars, machine learning (ML) classifiers. For this, two activity sequences, falling from a stationary position (FandS) and falling while standing up (WandF), from a publicly available dataset (N=15) is considered. Range-Time (RT), Range-Doppler (RD), and Doppler-Time (DT) maps were computed from radar signals. Shannon entropy features were extracted and classified using Random Forest (RF), Support Vector Machine (SVM), and NN with leave-one-out cross-validation. The proposed approach is able to discriminate elderly fall. For FandS, RF, SVM, and NN achieved F1 scores of 55.48%, 53.33%, and 61.27%, and Kappa coefficients of 0.24, 0.14, and 0.14, respectively. For WandF, F1 scores were 80.01%, 76.42%, and 47.10%, with Kappa coefficients of 0.55, 0.44, and -0.14. Thus, the proposed framework could be used for accurate detection of falls in smart homes.