PPL-net:Convolutional Neural Network-Based Polymorphic Points-Line verification framework
Wuqiang Cai, Runsheng Liu, Nan Chen, Yiming Luo, Yubin Zhou, Kaiqing Luo · 2023
To address the problems that the line segments extracted by the existing line segment detectors have local ambiguity and insufficient fitting accuracy, we propose PPL-net, a line segment detector based on the Modified Convolutional Neural Network (CNN) and Polymorphic Point-Line (PPL) verification framework. This method uses the point and line extraction algorithm based on CNN to obtain the initial line segment and feature point respectively. Then, the Point-Line verification framework consisting of feature point distribution detection, parallel over-density detection, and intersection over-extraction line segment detection is used to improve the line segment extraction performance. Using the Wireframe line segment data set, the PPL-net is tested and compared with other line segment extraction algorithms. Experimental results show that PPL-net performs better on average precision (AP), average recall (AR), F-measure and processing time (FPS), and has better disambiguation effect in dealing with line-segment dense scenarios.