Millimeter-Wave Radar-Based Elderly Fall Detection Fed by One-Dimensional Point Cloud and Doppler

This paper presents an elderly fall detection technique fed by one-dimensional (1-D) point cloud and doppler velocity. In the proposed technique, the long short-term memory (LSTM) network is created and then employed as an intelligent classifier of fall detection. 1-D point clouds and doppler velocity are the input data fed to the LSTM network. Experiments were conducted in order to verify the performance of the proposed fall detection technique. In the experiments, ten participants conducted various continuous sequence activities, for example standing still, walking, sitting, sleeping, simulating a fall, etc., in five different rooms. The millimeter-wave (mmWave) frequency-modulated continuous wave (FMCW) radar was employed to collect radar scattering signals that were transformed into data, including point clouds and doppler velocity. Different types of data grouped as inputs of the LSTM network of the fall detection system were investigated. The accuracy of the training and validation for the proposed system has shown that the point clouds in the
$z$
-axis direction and doppler velocity are adequate to be selected as the input data of the LSTM network. The proposed fall detection system can reduce overfitting problems and achieve the least number of input data features, resulting in the least computational complexity compared with state-of-the-art approaches. Before performing fall detection, the data were cleaned by using filtering, and the fault detection was reduced by using sliding window processing. After data preprocessing, the resulting outputs were employed for training and validation of the LSTM network. The window-size effect on the performance of fall detection using point clouds in the
$z$
-axis direction and doppler velocity was investigated, and the experimental results have shown that the proposed technique can detect a fall in real time. A fall detected by using the proposed system coincides with the activity of simulating a fall. The fall detection accuracy achieved by the proposed technique can reach up to 99.50%.