Fusing Wearable and Environmental Sensors for Context-Aware Firefighter Wellbeing Monitoring

Real-time monitoring of on-duty firefighters has the potential to reduce the risk of cardiovascular events and injuries by enabling early detection of fatigue and occupational hazards. Recent work uses wearable sensors to track metrics like heart rate and body temperature. However, most existing systems lack integrated environmental sensing, making it difficult to discern whether elevated vitals stem from physical exertion, ambient heat, or a combination of the two. To address this, a wireless sensing system is proposed that synchronizes physiological and environmental measurements during active firefighting. The system is validated using a new experiment where 28 firefighters completed repeated trials of a training circuit comprising warm-up activities and live-fire suppression. Participants answered survey questions after each phase to gauge overall wellness on a scale from 1 to 10. A predictive framework is developed to classify both the activity type and perceived wellness using the data collected. To extract useful information from the complex multimodal time histories, a series of models are compared using different sets of statistical features extracted from physiological, environmental, and demographic data. Furthermore, a linear mixed-effects (LME) model with subject-specific random intercepts is proposed to account for individual differences in biological response and self-reported wellness. Evaluation demonstrates the system differentiates warm-up from live-fire phases with over 97% accuracy. For wellness prediction, a one-label personalized LME model estimates their later ratings with a mean absolute error of 0.66±0.371, compared with 0.73±0.327 for the same model without personalization. Finally, data investigation reveals that wellness ratings were not fully explained by physiological measurements such as elevated vital signs, with individual differences and demographics accounting for most of the variance. These findings support further evaluation of multimodal sensing and personalized modeling for firefighter monitoring.