Pre-Impact Firefighter Fall Detection Using Machine Learning on the Edge
<- Publications
Pre-Impact Firefighter Fall Detection Using Machine Learning on the Edge
Xiaoqing Chai,
Boon-Giin Lee,
Matthew Pike,
Renjie Wu,
David Chieng,
Wan-Young Chung
IEEE Sensors Journal | 2023
| View on Publisher's Website
Abstract
Falling is a leading cause of firefighter casualties, making early fall detection critical for ensuring survival. Current fall detection systems (FDSs) struggle with real-time accuracy in complex firefighting environments. This study proposes a wearable pre-impact fall detection system (PIFDS) using machine learning (ML) and ensemble learning (EL) methods deployed on edge computing hardware to enhance real-time performance and accuracy. A moving thresholding method is introduced to address class imbalance in pre-impact phase data. Experimental results, based on data collected from 14 firefighters, demonstrate that a Decision Tree classifier with optimized parameters (DT-ED4) outperforms other ML and EL methods, achieving an average pre-impact detection lead time of 447.9 ms, with sensitivity of 95.10% and specificity of 97.99%. The system's heterogeneous IoT network architecture enables remote monitoring, allowing incident commanders to respond immediately to firefighter emergencies. This research advances wearable fall detection technology and extends its application to high-risk occupational environments.