A Method for Estimating Multiple Motion Parameters and Planar Surface Parameters without Feature Point Correspondence

A method estimating motion and planar surface parameters based on pixel intensity is proposed. Conventional approaches for estimation of these parameters are based on detected edge or feature points correspondence. Thus the problem is one of searching for any point which corresponds to feature points in other images. These approaches provide a solution from sparse points, however the shape cannot be estimated using all the pixels in an image. In this paper, by assuming that the multiple planar surfaces consist of objects, we propose a method that can estimate the depth of all pixels in input images without feature points correspondence. This approach places no restriction on the number of input images, which is different from conventional stereo vision with the exception of the factorization method. In the proposed method, parameters can be estimated as the ML (maximum likelihood) estimator, and the depth as the MAP (maximum a posteriori) estimator.