A Light Residual CNN for Radar Gesture Recognition

Gesture recognition based on radar technique offers unique advantages over traditional touch or camera-based methods for contact-less control, as the micro-Doppler signatures vary with changes in hand movements. In this article, a light residual convolutional neural network (CNN) is proposed for radar gesture recognition. Three residual modules of different structures are designed. A self-made dataset of $\mathbf{2 0 0 0}$ samples is collected using a commercial 24 GHz frequency modulated continuous wave (FMCW) radar module. It consists of time-frequency maps of nine hand movements and one empty gesture. Performance evaluation and ablation experiments are conducted on the dataset, and the metrics of precision, recall, accuracy, F<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> score, confusion matrix and running time are compared with four references. Experimental results indicate that the proposed LRCNN can accurately recognize gestures in real-time and achieve state-of-the-art performances. The source code and dataset are both uploaded on: https://github.com/Jiadongheng/gesture-recognition-matlab.git