Point Cloud-Based Through-the-Wall Human Activity Recognition Using IR-UWB Radar

In this paper, range-Doppler-time point clouds are generated using data captured by an impulse radio ultrawideband (IR-UWB) radar and a point cloud-based through-the-wall human activity recognition (HAR) network, called GLPointNet, is proposed. Specifically, the four-dimensional point clouds including range, Doppler, time, and intensity values are generated from IR-UWB radar echo signal. Subsequently, the point clouds are fed into GLPointNet, where global and local features are extracted simultaneously for recognition. The experimental results shown that the recognition accuracy of the proposed GLPpoint for the eight activities reaches 97.04%, while the network has the smallest number of parameters at 0.83M compared to other networks.