Computing Angle Of Rotation Based On Image Registration

Soil mechanics is an important branch of the engineering mechanics, it can provides the theory basis and practical method for geological process of soil, mainly used in Civil, Geology, Hydrology, road traffic. In the soil mechanics, the grains' action under the external conditions has a very important significance. In the experiment of the grains' deformation, in order to analyze the change of the structure and the movement, we need to detect and track grains accurately, on this basis we can measurement the displacement, rotation, deformation, etc. In allusion to the grains' rotation in soil mechanics experiments, we proposed two methods of image registration based on scale to calculate the rotation angle.This paper introduces the methods of image registration, and mainly analyzes the automatic registration algorithm; the method based on scale has higher noise immunity, so this paper use the method based on scale to calculate the rotation angle.There are two classic methods based on scale at present:Harris and SIFT. In this paper, we test these two methods at first, summarize the advantages and disadvantages, and then put forward more targeted matching strategy. Harris method is only about first order derivative: the angular points are uniformity and steady. But there's no characterization value for the Harris angular points, so we should provide other methods to calculate the rotation angle.The SIFT algorithm uses translation, rotation and affine transformation invariant of the image features in scale space and extracts image feature points. On this basis, we can calculate the rotation angle according to the keypoints' principal direction, and it'll be used in a follow-up data processing. When the key grain rotate, the background changes so little that we should get rid of the effect of the background keypoints.This paper experiment with different types of image sequence. The experimental data show that the method can get satisfactory results as Harris with an error about 10%, SIFT with an error about 5%.