Research On Line Matching Method Based On Relevant Points And Geometric Constraints

In the field of computer vision, image feature matching technology being applied to many fields, and makes the feature matching is more and more attention from the researchers, and its basic task is to use some features to associate with the two image under the same scene. Using the matching features mainly including point, line and area,which based on point feature matching has been widely studied, but the point feature descriptions of scene ability is weak, line features are widespread in the artificial scene,however, is the main information of this kind of scene could reflect the main buildings such as man-made scene contour edge character, therefore, line matching have become the preferred for image matching task under man-made scene, and is widely applied to target recognition, machine navigation and 3d reconstruction, etc. This paper focuses on line matching problem under the image transformation happened. In this paper, the main research contents and results are as follows:1. Several existing methods of line matching are introduced in detail, including methods based on texture in line neighborhood, methods based on geometrical information of lines’ and a combination of these methods. And analysis the advantages and disadvantages of various methods respectively.2. In this article, the basic and related theory of line matching will be described.The brief analysis of the concept of affine transformation and properties, and SIFT feature extraction principle was introduced in detail, a modified outlier removal method is proposed based on the scale-invariance to eliminate most SIFT matches, and through the experiments show that, in the image exists same or similar local texture structure cause false matching, this method can effectively eliminate false match. Line extraction is the precondition of line matching, using the line extracted by LSD algorithm for further processing, a combination of gradient amplitude line fitting method was proposed, it improve the precision of the line extraction, reduce linear fracture and discontinuous phenomenon.3. This paper proposes a line matching based on related points and geometric constraints. Firstly define SIFT feature point as related points, using related points and line construct the affine invariant to describe line, and match line by the similarity of affine invariant. Then calculation of affine transformation model, and finally obtained matched lines by the constraints of line angle and distance. This method not only usingthe local characteristics of lines, also considering the global geometric characteristics of lines. The experimental results show that, the method not only apply to image with affine transformation, and image with low texture, effectiveness of this method is demonstrated by experimental results.