Marker-based Extrinsic Calibration for Thermal-RGB Camera Pair with Different Calibration Board Materials

UAVs are used to rapidly and safely collect site data for engineering purposes. Thermal and RGB imagery are often captured for building envelope inspections. On large sites, the data produced can be cumbersome to process, especially if taken in a video format. This paper presents a way to perform image registration, or low-level data fusion, on thermal and RGB images that were simultaneously captured from rigidly fixed cameras - the same type as those found on commercial RGB-Thermal UAVs. The intention is to convert raw captured data into RGB-Thermal tensors that can be used in deep learning applications. Current techniques of geometrically calibrating a thermal camera can include complicated setups that make calibration at extreme angles difficult. Three calibration boards were involved in testing this image registration method: a cardboard and acrylic board, a wood and vinyl board, and a metal and vinyl board. Of the three, only the metal and vinyl calibration board yielded a consistent image, regardless of the time passed between the first and last captured image. This proves that calibration can be conducted cheaply and in any open area without obstructions.