AI-enabled human activity recognition: bridging contact-based and RF-based contactless sensing paradigms — a review

Human Activity Recognition (HAR) underpins applications in healthcare, security, smart environments and industrial safety. This survey examines HAR systems by contrasting contact-based sensing with radio-frequency (RF)-based contactless approaches, highlighting their respective detection mechanisms, performance trade-offs and available datasets. Contact-based methods that use accelerometers, gyroscopes, magnetometers and physiological sensors deliver high precision but face issues related to comfort, portability and long-term compliance. In contrast, RF-based contactless modalities, including WiFi channel state information, radar and radio-frequency identification enable scalable and privacy-preserving monitoring. However, they remain vulnerable to environmental noise, hardware limitations and multipath fading. Although contact-based sensing is discussed for comparative analysis, this review primarily focuses on artificial-intelligence (AI)-driven RF-based contactless HAR, analyzing sensing mechanisms, signal representations, publicly available datasets and associated learning paradigms. Covered applications span healthcare, security, gesture recognition, localization and smart homes. It also discusses challenges such as robustness, data scarcity, interpretability and ethics, concluding with future directions in multimodal sensing, edge efficiency, federated learning and explainable AI for transparent and reliable HAR systems.