Bi-directional Long Short-Term Memory Networks for Fall Detection using Bioradars

At present fall episodes present a huge threat to the health and life of the elderly. That is why a lack of effective and easy-to-use methods for automatic fall detection is a great problem of modern geriatrics. The present paper presents a method for unobtrusive fall detection of elderly based on a bioradar signal analysis. The method was tested using data of six volunteers performing different movement patterns in realistic surrounding conditions. The proposed long short-term memory network classifier proved to be applicable for detecting falls with an accuracy and Fl-score of 99 %.