Human Activity Classification With User-Held FMCW Radar Using Deep Recurrent Neural Networks

This letter proposes a human activity recognition method using user-held frequency-modulated continuous wave radar, which can be embedded in wireless devices, such as smartphones that come into direct contact with the user's body. Unlike the previous research that primarily measured human activities from a distance, our approach assumes that the radar is in the user's pocket, monitoring the movements of the body parts, such as limbs to recognize five distinct activities. We utilized range-Doppler maps to capture temporal changes in range and Doppler frequency, and employed deep learning models, including a 3D-convolutional neural networks (CNN) and a combination of long short-term memory (LSTM) with a 2D-CNN, to classify activities. Experimental results show that the LSTM-2D-CNN achieved a validation accuracy of 95.94%. Human activity classification using user-held radar offers robust performance while maintaining user privacy, making it suitable for a wide range of applications, such as healthcare, security, and defense.