Smart wearable IoT and multi-sensor fusion for firefighting

Abstract

Human Activity Recognition (HAR) algorithms have shown promise in firefighter risk assessment and behaviour monitoring. However, existing studies on HAR had not adequately addressed several critical challenges specific to firefighting scenarios. While various deep learning (DL) methods were proposed to increase classification accuracy, the more pressing needs in firefighting contexts were fast computation and timely notification. In addition, current HAR solutions were found to be unsuitable for recognizing the complex activities involved in firefighting. This thesis utilises multi-sensory fusion methods to explore advanced IoT-based wearable firefighter risk assessment system (FRAS) deployed with fall detection system (FDS) and firefighting activity recognition (FAR) algorithms. The research work consists of three associated studies that aim to answer the formulated research questions. The first study focuses on improving the accuracy and efficiency of a multi-IMU-based FDS using a novel attitude feature extraction (AFE) method. The second study introduces a pre-impact FDS (PI-FDS) model, which employed a dynamic thresholding method to tackle the issue of class imbalance. This study also investigate the feasibility of utilizing ensemble learning (EL) methods at the edge to increase real-time FAR performance. The last study presents a design and implementation of Internet of Things (IoT)-based wearable FAR system (IoT-FAR) for FAR remote monitoring. IoT-FAR explores the significance of adopting surface electromyography (sEMG), heart rate, and IMU features on the complex FAR. Results demonstrate that the proposed hybrid machine learning (HML)-based model outperforms existing state-of-the-art approaches with a mean accuracy of 98.29%. The IoT-based wearable FAR extends the application of HAR from healthcare to firefighting and serves as an inspiration for further research in other fields.

Date of Award	Nov 2024
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Boon Giin Lee (Supervisor), Matthew Pike (Supervisor) & David Chieng (Supervisor)