Deep Learning-Based Human Activity Recognition With FMCW Radar: A Review

Human activity recognition (HAR) has emerged as a critical research area with strong implications for healthcare, including elderly monitoring in assisted living and independent environments. Although several surveys have examined radar-based HAR, no comprehensive review has focused specifically on deep learning methods using frequency-modulated continuous-wave (FMCW) radar. To address this gap, we systematically analyzed 85 peer-reviewed publications spanning 2018-2025 from leading digital libraries. Our findings reveal a rapid growth in deep learning– enabled FMCW radar HAR, with major themes including activity classification, fall detection, and radar-based sensing for healthcare and IoT contexts. State-of-the-art models leverage convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders (AEs), and hybrid architectures to extract features from range–time, range–Doppler, micro-Doppler, range–angle, and point cloud domains. Despite notable progress, open challenges remain in computational complexity, limited public datasets, inter-class similarity, environmental robustness, and the absence of standardized evaluation frameworks. This survey synthesizes current advances and identifies research directions, providing practical guidance for researchers and practitioners developing next-generation FMCW radar–based HAR systems.