Keywords: thermal photogrammetry, RGB-IR picture fusion, visible and invisible data visualisation.

Conducting photogrammetry procedures on photos taken by thermal imaging cameras is extremely difficult or in many cases it is impossible. The reason for this is that even in the native images of relatively high-resolution thermal imaging cameras less information is stored by orders of magnitude compared to the native image of an RGB camera with a similar resolution. The procedures describing the pixels and their environment in this way cannot find point pairs between overlapping images or the quantity of the point pairs found is not sufficient for further processing. The intensity difference of the adjacent pixels of the thermal images is lower than as it is usual for colour images, thus gradient methods can be applied with low efficiency. In spite of all this, the analysis of an area’s orthophoto conducted within thermal range is more efficient and it is more informative than analyzing the same area’s discrete thermal images one by one. A method has been developed as well as functionally tested which with the combined application of a color and a thermal camera makes the production of a thermal orthophoto possible from aerial photos. As a result of the procedure, the orthophoto can be a conventional, grayscale image within the visible light spectrum, a false colour thermal image or a thermal image especially combined with visible light spectrum pixel data. The latter makes the picture, known by the human eye, visible together with the temperature data detected in the area, thus making a more accurate analysis and object determination possible. Using the method, such objects can be identified which individually are not distinct, neither in the native colour images nor in the native thermal images. The process does not work on an average or stronger PC, because the time it takes is not obviously calculatable. By utilizing the parallel computing capabilities of the appropriate hardware and GPUs, the processing time can be reduced to 100-130 hours depending on the expected accuracy. During the experiments, photogrammetric processing was performed using a Intel® Core™ i7-3820 Processor 32GB RAM with Nvidia Geforce GTX Titan Black machine. Further experiments are planned to test the combined use of multiple GPUs.

purchase the full-text of this paper (price £22)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £25 +P&P)