A planar homography estimation method based on multi-view total least squares

More and more computer vision methods are proposed based on planar homography. However, result accuracy of estimated parameters depends closely on homography estimation, which may be sensitive to noise. In this paper, a planar homography estimation method is introduced based on multi-view total least squares. It exploits both singular value decomposition and recursive technique in order to increase the computational accuracy of this algorithm. Applying this technique to parameter estimation would provide reliability and robustness in the present of noise since input perturbation of the system is considered. Simulations and experiments with real images validate this method.