Automatic Image Feature Matching Based on Reference Points

In computer vision and close-range photogrammetry, the result of 3D reconstruction depends on the quality of image matching. This paper presents an automatic matching algorithm for image feature based on reference points. This method uses the non-coded points and the coded points. The coded points matched easily are utilized for determination of distance relation between primitives. The non-coded points are utilized for 3D point reconstruction. Compared to previous methods, the novelties of the proposed method are as follows. Firstly, the proposed method improves automation level of algorithm by automatic detection using the reference points, which does not involve problem to choose manually point correspondences. Secondly, the 3D reconstruction precision is remarkably increased since the method of subpixel reference point center location is utilized. The experiment shows that the proposed algorithm can acquire both very high ratio of correct match and very high precision of 3D reconstruction