Automatic assessment of fuel load and fire risk via digitized database and intelligent computer vision

Abstract

Fuel load assessment is essential to evaluate fire hazard and risk in fire engineering design for infrastructure, safety management, and firefighting operations. This study introduces an intelligent method to automatically assess indoor fuel load and fire safety by leveraging a digitized fuel load database and computer vision. First, a well-trained fuel recognition AI model automatically estimates the fuel load through image segmentation and classification. Next, fire hazard is predicted based on a parametric temperature-time model to evaluate fire safety and risk. The AI-aided assessment tool is open-access in a web application for free and real-time operation by feeding images from surveillance cameras and 360 panoramic cameras. A case study in an open office demonstrates the smart fuel load assessment achieving an agreement of above 94%, compared to the digitized survey method. Based on the AI-predicted fuel load, the estimated fire duration and maximum gas temperature are 32% and 13% higher, respectively, than the code-based assessments. Moreover, a fire risk heatmap is auto-generated to visualize the spatial distribution of high-load fuels and potential fire spread hazards. This automatic method enhances the accessibility, convenience, and cost-effectiveness of fuel load assessment while ensuring commendable accuracy. The application of this AI tool enables more accurate predictions of fire behavior, thereby supporting smart firefighting strategies and more effective emergency response.