New Way to Look at Fire

Computer vision principles enable the analysis of fire, wind, and plume behavior from visual and infrared (IR) video as opposed to sparse measurements obtained with expensive instrumentation. Data that quantifies the transport of heat and fire spread, turbulent statistical information, and plume structure can be obtained from either visual or IR images and contribute to our evolving understanding of fire behavior. Unfortunately, black-box computer vision programs are not suitable due to the visually unique environment of fires and complex turbulent nature of their dynamics. I describe modifications of classical computer vision algorithms with adapted graph theory techniques that are applied to diverse instances of this environment and use them to extract data from prescribed fire videos. These data extraction experiments improve our understanding of the dynamics in complex environments and can validate fire spread models.