Multi-spectral Imaging System Calibration and Data Fusion for Applications in the Assessment of Rheumatoid Arthritis
Lens distortion as a result of the shape and construction of a photographic lens is a common problem in image acquisition. Thermal cameras are no exception to this artifact. So far many methods have been developed to formulate the distortion model and almost all of them exploit the patterns in visible range to calibrate the lenses in RGB cameras. A checkerboard is among the most common and well-defined patterns for RGB camera calibration. Unfortunately, most of those patterns will not be easily visible in images taken by a thermal camera. Furthermore, since the thermal cameras measure the infrared radiation (heat), the conductivity of the heat to the bordering objects in the pattern might mitigate sharp edges, which will make detection of relevant features within the pattern harder and less precise. In this paper we propose an algorithm to construct a calibration pattern visible for the thermal infrared cameras. We show how to extract robust features out of the sensed checkerboard pattern which is used afterward for lens distortion correction. Further, we evaluate our method and compare it to results obtained from well established algorithms for visible-light lens calibration. We also demonstrate how distortion correction improves the image registration between thermal and RGB aerial images taken by small-scale unmanned aerial vehicles (UAVs).