Thermal infrared video stabilization for aerial monitoring of active wildfires

Abstract

Measuring wildland fire behaviour is essential forfire science and fire management. Aerial thermal infrared (TIR)imaging provides outstanding opportunities to acquire suchinformation remotely. Variables such as fire rate of spread(ROS), fire radiative power (FRP) and fire line intensity maybe measured explicitly both in time and space, providing thenecessary data to study the response of fire behaviour to weather,vegetation, topography and firefighting efforts. However, raw TIRimagery acquired by Unmanned Aerial Vehicles (UAVs) requiresstabilization and georeferencing before any other processing canbe performed. Aerial video usually suffers from instabilitiesproduced by sensor movement. This problem is especially acutenear an active wildfire due to fire-generated turbulence. Fur-thermore, the nature of fire TIR video presents some specificchallenges that hinder robust inter-frame registration. There-fore, this paper presents a software-based video stabilizationalgorithm specifically designed for thermal infrared imageryof forest fires. After a comparative analysis of existing imageregistration algorithms, the KAZE feature-matching method wasselected and accompanied by pre- and post-processing modules.These included foreground histogram equalization and a multi-reference framework designed to increase the algorithm’s robust-ness in the presence of missing or faulty frames. Performanceof the proposed algorithm was validated in a total of ninevideo sequences acquired during field fire experiments. Theproposed algorithm yielded a registration accuracy between 10and 1000 times higher than other tested methods, returned 10xmore meaningful feature matches and proved robust in thepresence of faulty video frames. The ability to automaticallycancel camera movement for every frame in a video sequencesolves a key limitation in data processing pipelines and opensthe door to a number of systematic fire behaviour experimentalanalyses. Moreover, a completely automated process supports thedevelopment of decision support tools that can operate in realtime during an emergency.