High-Accuracy Fall Detection Using CW Radar Based on Free-Fall Signature

Falls are a major cause of accidental death and serious injury, particularly among the elderly. While various detection systems exist, wearable devices often suffer from low compliance, and vision-based methods raise significant privacy concerns. To address these issues, a radar-based fall detection system is proposed, which exploits free-fall signatures observed in both the time and frequency domains through a stepwise analysis. To isolate the rapid Doppler shifts characteristic of fall events, empirical mode decomposition (EMD) is applied to extract high-frequency components from the first intrinsic mode function (IMF1). Unlike voluntary daily movements that exhibit consistent and bounded frequency patterns, falls are involuntary and involve sudden free-fall motion, resulting in elevated power spectral density (PSD) in higher-frequency bands that are rarely observed in ordinary activities. The proposed algorithm identifies fall events by measuring the PSD from IMF1in the band above 50 Hz and discriminating them from non-fall motions. Experimental validation using an in-house 5.8-GHz continuous-wave radar shows that fall events exhibit, on average, a PSD increase of about 17.5 dB compared with typical daily activities. A total of more than 1,000 experimental trials were conducted with five volunteers possessing diverse physical profiles (height: 158–180 cm, weight: 50–93 kg), verifying the reliability of the proposed system and demonstrating its robustness across different body characteristics.