Human Fall Detection Based on Machine Learning Using a THz Radar System

A terahertz (THz) frequency modulated continuous wave radar system was used for human fall detection in this paper. Five features were extracted from the range-time spectrograms and time-frequency spectrograms for further classification. Eight machine learning classifiers have been implemented including the logistic regression, the support vector machine, the k-nearest neighbor, the decision tree, the Naive Bayes, the quadratic discriminant analysis, the adaptive boosting and the back propagation neural network. Different kinds of combinations of five features were tested to obtain the optimum combination when 10 cross-validation and 100 repeated time tests were considered. In the experiment, total 600 motions including 300 falls and 300 non-falls from 10 different subjects were acquired by the THz radar system in a normal room. Eight machine learning classifiers all showed a good performance with AUC values larger than 0.921 when detecting falls.