With the growth of the number of elderly people, fall detection based on radio frequency signals for those vulnerable groups emerged. In radar-based fall detection methods, different heatmaps are generated from the ec...
详细信息
With the growth of the number of elderly people, fall detection based on radio frequency signals for those vulnerable groups emerged. In radar-based fall detection methods, different heatmaps are generated from the echo signal. However, the noise and interference existing in the environment have a significant impact on heatmaps, which eventually lead to bad performances in fall detection. In this article, an anti-fixed-interference algorithm based on range-Doppler maps cancellation and a denoising algorithm based on Rayleigh probability distribution are proposed to ensure the consistency of the target's Doppler information in a Doppler-time (DT) map. To leverage the micro-Doppler information extracted from the movements of human limbs during falls, we construct a DT-range (DTR) map, which fuses 3-D information in a single map. Moreover, a neural network composed of 3-D convolutional neural networks and a bidirectional long short-term memory network is designed to extract 3-D features in DTR maps. The experiment is conducted in three different scenes, where a giant metal chassis, pipe vibration, and a fan are presented as an interference, respectively. Finally, based on the proposed methods, we achieve 96.16%, 94.21%, and 93.06% precision of fall detection in the above scenes, respectively.
暂无评论