Motion estimation based on axis affine model
Zhihui Li, Chunyan Shao, Yongmei Liu · 2010
Motion estimation is an important component of video processing system. Affine model with 6 parameters is a common motion model. A new affine model - axis affine model (AAM) which represents the affine motion by main axes of a rectangular or ellipse region is presented in this paper. The original abstract model is changed into the form of the compounding of several simple transformations. The affine parameters become the shift of main axes, rotation angles and scale coefficients which are much clearer in geometry than the traditional parameters and easier in their range determination. The application of axis affine model makes it possible to estimate the motions by main axes, which accelerates the estimation while keeping the same degree of accuracy. In this paper the axis affine model, objective function of robust error and DIRECT algorithm are used to implement the motion estimation. The experiment results prove the effectiveness of the approach in this paper.