<p>Underground fires create smoke-filled environments, where poor visibility is a primary cause of fatalities. To address this, this study presents a smart helmet prototype designed to restore visibility and perception enhancement for mine rescuers. The system integrates a thermal imaging camera, a Raspberry Pi 5 processing unit, and an augmented-reality (AR) display to provide a hands-free, real-time thermal view when visual light is obscured. A prototype was developed to operate fully offline to address unreliable underground connectivity. Visibility was tested in a 19ft x 7ft x 7ft container filled with dense fog, simulating zero-visibility conditions. The methodological approach adopted the EfficientDet-Lite0 machine-learning model for thermal-based human detection on embedded hardware. Testing was performed twice, with heaters as targets, and once with three people as targets. A data-clipping technique was applied to normalize the thermal feed and enhance human-target contrast. The thermal camera clearly identified targets through dense fog where visible light failed. The Raspberry Pi 5 maintained real-time processing (approx. 20 FPS in fog, 12 FPS with detection). The hands-free AR display proved effective for maintaining mobility. Despite not being retrained, the AI model achieved a moderate confidence output despite the domain shift, with an average detection score of 0.718 under a controlled environment, confirming the approach’s feasibility. This study successfully demonstrates that a wearable, offline system combining thermal imaging, edge AI, and AR can restore vision in zero-visibility conditions. The prototype confirms the viability of adapting RGB-based models for thermal detection but highlights the critical need to retrain domain-specific thermal datasets to improve accuracy.</p>