Abstract
Falls are a leading cause of mortality among individuals aged 65 and above, making timely fall detection alarms essential for preventing fatalities. Contactless radio-frequency (RF) technology for fall detection has gained traction due to its wide coverage and privacy-preserving features. However, existing RF-based systems often assume falls create predictable RF signal patterns, which can be problematic, especially among visually impaired people (VIP), who cannot detect environmental changes. To overcome this challenge, we propose an innovative approach that focuses on recognising normal, repeatable human activities and detecting falls as deviations from these patterns. Our prototype, developed using commercial UWB Xethru radar, was tested on human subjects, including VIPs. The results demonstrated a classification accuracy of 98.8% within a 1.5-meter range in indoor environments, proving our system’s high reliability and adaptability for real-time fall detection. This approach provides a more dependable solution for protecting the elderly, especially those with visual impairments, from fall-related dangers.
Original language | English |
---|---|
Journal | IEEE Sensors Journal |
Early online date | 9 Dec 2024 |
DOIs | |
Publication status | E-pub ahead of print - 9 Dec 2024 |
Keywords
- visually impaired children
- contactless fall detection
- real time detection
- machine learning