Multimodal registration of high-resolution thermal image mosaics for the non-destructive evaluation of structures

This paper proposes an approach to automatically register thermal images to visible images by utilizing known relative camera positions. There are multiple areas of application for the fusion of visible and thermal images, but previous approaches require feature detection and matching in both visible and infrared ranges in order to register the two imaging modalities. In the application domain of structural diagnostics, it is not guaranteed that corresponding features will be present in the thermal images. The diagnostic imaging is being performed on building-scale targets which is another characteristic of this application that separates itself from prior thermal-visible registration studies. Non-destructive evaluation of structures could require the inspection of entire building facades which necessitates many thermal images to yield a desirable level of resolution and detail. The acquisition of high-resolution thermal image mosaics generates the level of detail required and provides context for the entire target structure. A low-cost camera mount was custom designed and 3D printed in order to mount visible web cameras around a thermal camera. Once the system is calibrated with a pair of multimodal images that contain corresponding features, every subsequent pair of images can be automatically registered. An application for this approach is presented in which the generation of a visible image mosaic in turn automatically generates a registered thermal image mosaic. The benefit of this approach is that there is no reliance on features being present in the thermal images in order to register them to the visible images.