Line Matching And 3D Modeling Using Geometric Invariants

Feature matching is a very important issue in computer vision problems.Line feature is one of the most fundamental features in image matching problems,which is widely used in many applications including 3D modeling,a popular topic in the field of computer vision.We propose a novel line matching method based on a newly developed projective invariant namely characteristic number.The interest point matching results are taken to lead the works of line matching.We construct the line-points invariants based on the points in the neighborhoods of each line and the similarities between the line neighborhoods can be computed through the invariants.As the property of the invariants suggest that the points in the neighborhoods with high similarity are possibly lying on the same plane surface,we can get the transformations between such planes,which can filter the potentially overlapped line pairs in two images.Finally,a weighted voting strategy is used to disambiguation and obtain the matched lines.Experiments show that the proposed method is robust to various image changes,and can get better performance than some state-of-the-art methods especially in wide baseline images.Most of the existing methods are based on the interest point matches or the line matches.The point-based methods are more mature,but usually less intuitive.Lines can provide more structural information and thus more intuitive,however,lines have more complex geometric properties,which makes the methods line-based methods usually time consuming.In this paper,we propose a simple and effective approach of 3D modeling.We first do point matching and line matching with existing methods,then we take the intersections of the real lines and the virtual lines passing through interest points to construct the 3D models.As the points and lines are matched,we can get the correspondence of intersections easily.A simple invariant is taken to filter out fake intersections and wrong matches.Because the intersections are located on real lines,we can use the mature technology of point-based methods while the3 D point cloud maintain the structural information of lines.Experiments show that we only need a few number of interest points to get 3D models,so the translation from lines to intersections is very fast,and the 3D contain more meaningful information than traditional point based methods.