False matches filtering methods for binocular vision

Outliers (false matches) detection techniques are very important for computer stereo vision. There always exist some outliers in spare matching and dense matching which impact the following 3D reconstruction in effect and accuracy. The author discussed some detection algorithms for outliers on the basis of 2D space and 3D space, and put forward a grey pre-processing algorithm applying beforehand to lower the false matching rate. Meanwhile, a kind of disparity filtering algorithm based on ground truth control points is also put forward. The former only balances the overlap area between two images which can reduce the calculation quantity; the latter based on disparity mean value filtering can improve the detection efficiency. Experiment results of real images proved that the false matches can be detected and eliminated effectively by the new algorithms which ensure the effect and accuracy of 3D reconstruction.