NPConvNet: A Lightweight Network for Gesture Recognition Based on Millimeter-Wave Radar
Wenxin Xiong, Sha Huan, Xiaoxuan Yang · 2024
Radar gesture recognition provides a more flexible and intuitive way of non-contact human-machine interaction in smart homes and intelligent vehicles. Existing methods have insufficient consideration for the lightweighting of the recognition algorithms or networks in embedded applications. In this paper, we propose a new partial convolution (NPConv) by changing PConv forward fusion to reverse fusion that achieves higher accuracy compared to PConv. Building upon our NPConv and Depthwise Separable Convolution, we further propose NPConvNet, whilch utilizes preprocessed radar Range Time Maps (RTM) and Doppler Time Maps (DTM) to identify different hand gestures. And the NPConvNet achieves lightweighting while ensuring the performance. Experimental results validate the superiority of this network compared to traditional networks.