Computer vision methods for automating high temperature steel section sizing in thermal images

This paper proposes a solution to autonomously measuring steel sections with images captured by a monocular, uncalibrated thermal camera. A fast structural random forest algorithm extracts the edges of the steel sections from sequentially coming image data. Two approaches are proposed that recognize the edges and remotely evaluate the size of the manufacturing objects of interest, which will facilitate automating the steel manufacturing process. Four sets of experiments are conducted, and the results show that our method achieves accurate dimension measuring results, with a root mean square error less than 2:5 mm, which is the maximum tolerance bound of the manufacturing process.