LAUSR

Abstract Image registration is the major problem to many applications including measurement in the area of computer vision. In this paper, a scheme of high-precision measurement based on the stereo camera system is proposed. To take advantages of the speed up robust features and the optical flow features, they are chosen to be the registration basis for estimating and refining the initial geometrical transformation, This paper reviews the existing methods in the optimization procedure of image registration. By analyzing the weakness and strength recent progress in global polynomial optimization, a hybrid method which combines the merits of the SVD method, interleaving parameterization algorithm and the convex linear matrix inequalities relaxation method together is presented to achieve the global solution of the fundamental matrix and the rotation matrix problems. The experimental results confirms the high precision and stability of the proposed method.