Aligned Anchor Groups Guided Line Segment Detector

TL;DR

This work introduces AAGLSD, a line segment detector that leverages hierarchical anchors—regular anchors and aligned anchor groups (AAGs)—to produce complete, precise line segments. By framing detection as an anchor-linking process guided by gradient magnitude and level-line alignment, and using a simple validation and merging step, the method achieves strong precision-recall performance while remaining efficient. Key contributions include the hierarchical anchor extraction, the pixel-routing based linking strategy with skipping, and a straightforward validation/merging scheme that reduces false positives and duplicate predictions. Experiments on YorkUrbanDB, YorkUrban-LineSegment, and HPatches demonstrate competitive or superior performance among handcrafted detectors and notable robustness to illumination changes, highlighting practical applicability in structured-scenes tasks.

Abstract

This paper introduces a novel line segment detector, the Aligned Anchor Groups guided Line Segment Detector (AAGLSD), designed to detect line segments from images with high precision and completeness. The algorithm employs a hierarchical approach to extract candidate pixels with different saliency levels, including regular anchors and aligned anchor groups. AAGLSD initiates from these aligned anchor groups, sequentially linking anchors and updating the currently predicted line segment simultaneously. The final predictions are derived through straightforward validation and merging of adjacent line segments, avoiding complex refinement strategies. AAGLSD is evaluated on various datasets and quantitative experiments demonstrate that the proposed method can effectively extract complete line segments from input images compared to other advanced line segment detectors. The implementation is available at https://github.com/zyl0609/AAGLSD.

Figures (8)

• Figure 1: The search regions for pixel-routing, corresponding to various linking directions and the orientations of the fitting line. Blue grids indicate the candidate pixels associated with the $forward$ direction (red arrow), and green grids are associated with the $backward$ direction (orange arrow).
• Figure 2: The linking process links AAGs progressively and form a line segment, where every three pixels with the same color is an AAG.
• Figure 4: Parameters between the prediction and the ground truth. $d_a$ is the angle difference; $d_c^{pred}$ is the perpendicular distance from the midpoint of the predicted segment to the ground truth, while $d_c^{gt}$ is the opposite; $d_e^1$ and $d_e^2$ are the distances between the endpoints; $len_{gt}$ is the length of the ground truth; $len_{inter}$ is the length of the projection of predicted line segment onto the ground truth.
• Figure 5: Performance curves for line segment detectors evaluated on the YorkUrban datasetcoughlan2003manhattan and the YorkUrban-LineSegment datasetcho2017novel in various $\lambda_{area}$.
• Figure : (a)