Line matching plays an essential role in Structure from Motion (SFM) and Simultaneous Localization and Mapping (SLAM), especially in low-texture scenes, where feature points are hard to be detected. In this paper, we present a new method by combining Convolutional Neural Networks and Graph Convolutional Networks to match line segments in pairs of images. We design a graph-based method to predict the assignment matrix of two feature sets with solving a relaxed optimal transport problem. In contrast to handcrafted line matching algorithms, our approach learns the line segment features and performs matching simultaneously through end-to-end weakly supervised training. The experiment results show that our method outperforms the state-of-the-art techniques and is robust to various image transformations. Besides, the generalization experiment illustrates that our method has good generalization ability without fine-tuning. The code of our work is available at  https://github.com/mameng1/GraphLineMatching.

线匹配在“运动结构（SFM）”和“同时定位和映射”（SLAM）中起着至关重要的作用，尤其是在难以检测特征点的低纹理场景中。在本文中，我们提出了一种通过结合卷积神经网络和图卷积网络来匹配成对图像中的线段的新方法。我们设计了一种基于图的方法来预测两个特征集的分配矩阵，并解决一个宽松的最优运输问题。与手工的线匹配算法相比，我们的方法学习线段特征，并通过端到端的弱监督训练同时执行匹配。实验结果表明，我们的方法优于最新技术，并且对各种图像转换均具有鲁棒性。除了，泛化实验表明，该方法具有良好的泛化能力，无需进行微调。我们的工作代码可在以下位置获得：  https://github.com/mameng1/GraphLineMatching。