Automatic fall detection by using Doppler-radar and LSTM-based recurrent neural network

Abstract

Falls are a major public health concern among seniors living independently at home. This article presents a new approach for automated real-time fall detection by a single continuous-wave micro-Doppler radar and deep learning. Unlike related methods, we apply Long Short-Term Memory based recurrent neural network (RNN) to time-frequency spectrograms of Doppler-radar returns to identify falls. The approach does not require extra hardware or big data set to classify falls from non-fall actions and has superior accuracy compared to CNN-based fall detection.

Topics & Concepts