Dual mmWave radars fall detection using the Stockwell transform

Abstract

Elderly fall detection based on radar technology has recently drawn more and more attention because of the advantages of privacy protection, all-day working, and non-intrusive properties. Numerous time-frequency representation methods have been used to extract fall signatures. However, the measurement of peak-peak Doppler frequency using short-time Fourier transform and wavelet transform is limited due to weak energy in the high Doppler frequency area. A method for human fall detection based on Stockwell transform with dual mm Wave radars is proposed in this paper. Two mmWave radars are vertically placed to collect human motions in different directions, and the Micro-doppler features of human motions are obtained using Stockwell transform. To enhance the feature for recognition, extracted ridgelines from two mm Wave radars are fused in the time-frequency domain. And the model trained by a convolutional neural network (CNN) is used for fall recognition. 960 fall motions and 720 non-fall motions from 60 volunteers were collected, and the experimental result shows that the proposed method can produce a recognition accuracy of up to 94.14%.