Improved algorithm for line segment matching based on reweighted random walk

In this paper, we make improvements on the existing line segment matching pipeline which is mainly based on random walk algorithm and local homography constraint. In terms of algorithm efficiency, we introduce a topological descriptor to narrow the searching scope and guide the restoration from line group matches to line segment matches. To speed up the free line segment match supplement, a fast free match search method is also proposed in this paper. In terms of the matching quality, we propose a bidirectional multiple selection algorithm, which groups collinear line segments into clusters and determines whether to select the corresponding matches based on the confidence of the clusters. To ensure the precision of the matching algorithm, we add a conflict match filtering mechanism in the final supplemental stage. Experimentally, it is proved that the line segment matching algorithm proposed in this paper has a high recall rate and faster speed while guaranteeing precision.