Property Similarity Line Segment Detector
Juncan Deng, Jintao Cheng, Weibo Cai, Yubin Zhou, Rui Fan, Kaiqing Luo · 2021
Most line segment detection methods suffer from over-segmentation. Therefore, we propose a line segment detector based on property similarity. It is composed of three steps. First, after the original line segments are generated, the competitive grouping method is used to group line segments with many same aligned endpoints, and a minimum bounding rectangle is adopted to enclose every group. In the second step, the line segments in the same group are verified whether they have similar positions and gradients. Finally, the center lines of the rectangles are the final line segments. We show that our method can significantly detect complete line segments and much less prone to over-segmentation while maintaining real-time performance. The experimental results illustrate that our method achieves a maximum F-score of approximately 27%, which about 5 % more than other leading methods when analyzing the images from the York urban line segments dataset.