An Enhanced Human Fall Detection System Using mmWave Sensors for Indoor Smart Space

This paper introduces an enhanced mmWave human fall detection (EMFD) system that uses sequential millimeter-wave(mmWave) radar frames to improve the performance of fall detection. The approach is designed to enhance smart healthcare for elders and patients who may be at risk of falling in areas where cameras are not suitable. By integrating privacy-friendly mmWave technology with enhanced variational recurrent auto-encoders (EVRAE), the system enables the smart space to detect falls even in privacy-sensitive space such as bathrooms and fitting rooms. Recent mmWave fall detection usually employ the body benchmark drops in vertical direction causing the spike in anomaly level. However, simply relying on the threshold of body centroid height dropping may lead to misdetection of falls in complex scenes that include various normal activities of daily living (ADL), such as sitting, squatting, and stooping, which can vary in different countries. The proposed system chooses a head point above the body centroid as the benchmark, instead of only using the body centroid as in other mmWave fall detection systems. Utilizing a dataset from both experimental and commercial environments, this system is trained on ADL. Utilizing a dataset from both experimental and commercial environments, this system is trained on normal human ADL. The system detects abnormal patterns, such as falls, identified as spikes by the EVRAE during the inference stage. The results demonstrate that the proposed system may effectively handles complex sensory data in smart environments and provides a privacy-preserving approach to enhancing user experience and healthcare management, particularly in the context of elderly fall prevention.