ELSD: Efficient Line Segment Detector and Descriptor

We present the novel Efficient Line Segment Detector and Descriptor (ELSD) tosimultaneously detect line segments and extract their descriptors in an image.Unlike the traditional pipelines that conduct detection and descriptionseparately, ELSD utilizes a shared feature extractor for both detection anddescription, to provide the essential line features to the higher-level taskslike SLAM and image matching in real time. First, we design the one-stagecompact model, and propose to use the mid-point, angle and length as theminimal representation of line segment, which also guarantees thecenter-symmetry. The non-centerness suppression is proposed to filter out thefragmented line segments caused by lines' intersections. The fine offsetprediction is designed to refine the mid-point localization. Second, the linedescriptor branch is integrated with the detector branch, and the two branchesare jointly trained in an end-to-end manner. In the experiments, the proposedELSD achieves the state-of-the-art performance on the Wireframe dataset andYorkUrban dataset, in both accuracy and efficiency. The line descriptionability of ELSD also outperforms the previous works on the line matching task.