The Research Of Global Motion Estimation Algorithm Based On SIFT

Global motion estimation is one of the basic methods for computer vision and video processing. It has been applied widely in medical image analysis, robot vision navigation, mobile target tracking, video compression coding and many other fields. The research about it has important theoretic meaning and highly practical value. Scale invariant feature transform descriptor is a kind of local feature representation with high robust performance. Since it was been proposed, it has been focused by more and more researchers.At first, we analyze the present methods of global motion estimation, and introduce motion parameter model which is used in the algorithm of this paper. Based on the analysis to the predecessor's fruit, indicate the contradiction between estimation accuracy and computational complexity. Then, utilizing the invariance of SIFT descriptor to scale and rotation, combing with motion parameter model, apply SIFT descriptor to global motion estimation. The general idea of the algorithm in this paper is, obtain spatial coordinate of the feature points and these corresponding relation based on SIFT matching; then compute model parameters of six affine motion model by least square method. To eliminate the influence of mismatches and noise, we propose two kinds of methods based on image content and motion vector.Through experiment on real image, the result show that the ideal is reasonable, and the methods to eliminate mismatches and noise are effective. The effect of global motion estimation is satisfactory.