Fail detection problem for smart home-care systems using an ultra wideband radar is considered in this paper. The goal is to identify the occurrence of fall from the radar return signals through a supervised learning approach. To this end, a new framework is proposed based on stacked long-short-term memory (LSTM) recurrent neural network to develop a robust method for feature extraction and classification of radar data of human daily activity. It is noted that the proposed method do not require heavy preprocessing on the data or feature engineering. It is known that LSTM networks are capable of capturing dependencies in time series data. In view of this, the radar time series data are directly fed into a stacked LSTM network for automatic feature extraction. Experiments are conducted on radar data collected from different subjects, when performing fall and non-fall activities. It is shown that the proposed method can provide a classification accuracy higher than that yielded by the other existing methods.

Additional Metadata
Keywords Fall Detection, Smart Home Care, Stacked Lstm-Rnn, Time Series, Ultra Wideband Radar
Persistent URL dx.doi.org/10.1109/LSC.2018.8572048
Conference 2018 IEEE Life Sciences Conference, LSC 2018
Citation
Sadreazami, H. (Hamidreza), Bolic, M. (Miodrag), & Rajan, S. (2018). On the use of ultra wideband radar and stacked lstm-rnn for at home fall detection. In 2018 IEEE Life Sciences Conference, LSC 2018 (pp. 255–258). doi:10.1109/LSC.2018.8572048