Robust Line Matching for Image Sequences Based on Point Correspondences and Line Mapping

Abstract

This paper proposes a line segment matching method by performing line mapping and unmapping based on point correspondences. The goal of this paper is to improve the accuracy and the robustness of line segment matching for two views, which will be conducive for generating a full singleline structure for image sequences. In this paper, to improve the quantity and quality of line matches, the topological adjacency of a point-line is first introduced for two goals: to find and filter the candidate line segment by resorting to KD-tree data-index structure efficiently in the corresponding image and to leverage the candidate line segment to improve the performance of planar homography. In addition, the shift threshold parameter is theoretically analyzed, and trials are validated to determine the matching degree. Line mapping and unmapping are then used for image sequences to reduce missing matches. The extensive experimental results validate and demonstrate that our method is both more accurate and robust than existing line matching methods for two views under the circumstance of a higher recall rate. In addition, our method contributes in finding full line segments for image sequences using line mapping, with higher completeness of the 3D line model than that obtained by the state-of-the-art methods. INDEX TERMS Line mapping, line segment matching, planar homography, topological adjacency.

Research on 3D Modeling of Wupaolong based on Sparse Point Cloud Reconstruction of SFM Algorithm

“…In formula (1), S is the collection of pixel coordinate points in the neighborhood of (x, y) points, M is the total number of pixels in the neighborhood, f (x, y) is the pixel of the original image, and g (x, y) is the pixel after the filter template is processed. The advantage of the mean filtering algorithm is that it is simple and easy to operate, but the disadvantage is that it is easy to lose the edge information of the image [14].…”

“…If they are not coplanar, they are directly removed from the remaining searchable set, and the judgment of the next facet is continued. It can be seen from formula (14) that the lower the sample data error rate, the higher the probability of obtaining at least a subset composed of interior points, and the fewer sampling times required. After filtering, the rough points in the point cloud have been eliminated, and the triangulation network constructed by it can be regarded as a sample with a low data error rate.…”