A New Robust Bayesian Method for the Affine F-Matrix Estimation.
Sami S. Brandt, Jukka Heikkonen · 2000
In this paper, a new robust method is proposed for the ane fundamental matrix estimation problem in stereo vision. The method is based on the weighting of the interest points by their a posteriori probabilities to be relevant. Not only being robust against false matches the proposed method generally poses a new aspect for outlier rejection which has applications on a wide range of elds. The method is evaluated with both synthetic and real data. 1 Introduction Matching two images of a single scene is one of the basic problems in computer vision. Stereo, motion, and 3D object recognition are all multiple view problems sharing the same geometrical constraint, the epipolar constraint [9]. However, when the motion between the two images is not known the corresponding epipolar geometry is also unknown. The fundamental matrix consist of the minimum amount of information between two images in order to construct the epipolar geometry between them. The importance of the fundamental matrix ha...