Huang Siyu ; Qin Fangbo ; Xiong Pengfei ; Ding Ning ; He Yijia ; Liu Xiao

Abstract

This paper proposes a novel deep convolutional model, Tri-Points Based LineSegment Detector (TP-LSD), to detect line segments in an image at real-timespeed. The previous related methods typically use the two-step strategy,relying on either heuristic post-process or extra classifier. To realizeone-step detection with a faster and more compact model, we introduce thetri-points representation, converting the line segment detection to theend-to-end prediction of a root-point and two endpoints for each line segment.TP-LSD has two branches: tri-points extraction branch and line segmentationbranch. The former predicts the heat map of root-points and the twodisplacement maps of endpoints. The latter segments the pixels on straightlines out from background. Moreover, the line segmentation map is reused in thefirst branch as structural prior. We propose an additional novel evaluationmetric and evaluate our method on Wireframe and YorkUrban datasets,demonstrating not only the competitive accuracy compared to the most recentmethods, but also the real-time run speed up to 78 FPS with the $320\times 320$input.