A Fast and Efficient Lines Matching Method via Multi-depth-layer Strategy

Lines matching is the significant image pre-processing technique, which plays a central role in 3D reconstruction, visual navigation and other research fields. However, traditional lines matching methods suffered due to issues, e.g., complex processes, low efficiency, and poor matching effect, while those drawbacks strongly hurt the performance as required in the V-SLAM. In this research, we propose a fast and effective lines matching method. Based on the previous research of the fast line detection, we make full use of depth information to construct line features candidate areas to eliminate invalid features and to reduce the computational cost. Then, we use LBD descriptor to inscribe line features, and thereby ensuring the proper lines matching. It is worth noting that, in searching the effectiveness as required by tasks of lines detection and matching, we introduce geometric constraints into our framework. Experiments show that the method proposed in this paper can effectively improve the effectiveness and efficiency of the lines matching in real V-SLAM tasks.