Homography Estimation Based on Order-Preserving Constraint and Similarity Measurement

Abstract

Homography is an important concept that has been extensively applied in many computer vision applications. However, accurate estimation of the homography is still a challenging problem. The classical approaches for robust estimation of the homography are all based on the iterative RANSAC framework. In this paper, we explore the problem from a new perspective by finding four point correspondences between two images given a set of point correspondences. The approach is achieved by means of an order-preserving constraint and a similarity measurement of the quadrilateral formed by the four points. The proposed method is computationally efficient as it requires much less iterations than the RANSAC algorithm. But this method is designed for small camera motions between consecutive frames in video sequences. Extensive evaluations on both synthetic data and real images have been performed to validate the effectiveness and accuracy of the proposed approach. In the synthetic experiments, we investigated and compared the accuracy of three types of methods and the influence of the proportion of outliers and the level of noise for homography estimation. We also analyzed the computational cost of the proposed method and compared our method with the state-of-the-art approaches in real image experiments. The experimental results show that the proposed method is more robust than the RANSAC algorithm.