This paper proposes an adaptive method for dual camera based steel section sizing, where high accuracy measuring is challenging due to the lack of well pronounced image features. The proposed approach includes additional information from a sidewise positioned checkerboard and enables adaptive image registration. A thorough evaluation of the registration results based on the virtual checkerboard is presented. On the accomplishment of image registration, both a fast Fourier transform and a discrete wavelet transform are adopted for fusion of the registered images. A range of comparisons with various metrics is conducted to achieve the best fusion quality. The hot steel section sizing results show an accuracy that is in line with the rolling standards, i.e. in the tolerance range less than 1.5 mm error.