Robust point matching method by vector structures

A novel robust point matching method is presented in this paper, which can find corresponding point pairs effectively via 2D or 3D points' relative positions. There the relative positions of point sets which come from the same moving rigid object are invariable. To describe the relative position, a vector structure is constructed by connecting one point to the rest. If two points in different sets is a matching pair, their vector structure should be similar, in which the corresponding vectors will have approximate lengths and mutual angles. Our method is based on this kind of similarity. Firstly we apply an algorithm based on Hausdorff Distance to choose the candidate matching pairs for computation reduction. To compare the two structures' similarity, both lengths and mutual angles of vectors should be considered. By finding a benchmark vector, we acquire mutual angles of vectors in the structure; then a similar coefficient is presented to determine if the two structures is a matching pair. After calculating all the similar coefficients of the candidate pairs, a coefficient matrix is built. We give a search strategy to elect matching pairs from the matrix. Through the experiments, it is proved that this method can keep low mismatching ratio even when there are quite large position errors and lots of disturbing points. The method can be applied well to two or three dimensional situation.