The development of a passive Wi-Fi radar fall detection system for elderly based on cumulative sum thresholding

Falls among elderly individuals aged 60 and above create significant health risks due to age-related physical and medical impairments. These incidents often result in severe injuries or fatalities. Although various fall detection systems have been developed, existing methods have some limitations. Wearable sensors require consistent user compliance which can be inconvenient, while camera-based systems raise privacy concerns and demand high computational power. Radar-based solutions have emerged as promising alternatives due to their ability to detect and classify fall events with high accuracy. This study proposes a Passive Wi-Fi Radar (PWR)-based fall detection system that leverages existing in-home Wi-Fi signals. Key challenges include ensuring reliable signal propagation, identifying sensor placement, and differentiating falls from normal activities in multi-occupant settings. Furthermore, most radar-based systems rely on extensive post-processing, which limits their suitability for real-time applications. Hence, this study proposes a threshold-based approach using the Cumulative Sum (CUSUM) method to enable real-time fall detection. The research commences with the simulation of Wi-Fi signal propagation for determining receiver placement to enhance coverage for human motion detection. Hardware components were subsequently tested and validated to ensure optimal signal reception. The study further investigates fall detection in single and dual-activity scenarios involving multiple occupants, addressing the inherent challenges of distinguishing overlapping motion signals. Experimental validation was conducted within a controlled indoorenvironment under ethical approval, where trained stunt performers executed fall simulations while volunteers performed non-fall activities to refine and improve detection accuracy. The results demonstrate that the proposed PWR-based system utilising the CUSUM time-based method delivers a reliable solution, with detection accuracies of 95.83% for single falls and 93.75% for dual activities. This research advances non-wearable fall detection technologies by offering a privacy-preserving alternative to existing solutions. Future work will focus on optimising system performance and deploying the technology in real-world environments to enhance adaptability and robustness.