Abstract:

 We propose an algorithm for square grid image matching to be applied to the measurement of steel forming strain. The thinned image of grid lines is obtained with image binarization and thinning. A coarse-to-fine algorithm is executed to extract the sub-pixel image coordinates of grid nodes and establish topological relationship between nodes. An automatic matching algorithm is proposed to establish correspondence among grid nodes in multiple images according to the topological relation and epipolar constraint based on the known camera position and orientation in each shot. Experimental results show that the average epipolar distance of all
matching points is approximately 0.17 pixels, and the matching method is stable and reliable. The results can be used in the subsequent strain calculation.

Key words: sub-pixel matching, strain measurement, grid image, node extraction, 3D reconstruction