Answers are grounded in this paper and cite their sources.
Thin overhead wires of approximately 1 to 3 cm are difficult to preserve in Unmanned Aerial Vehicle (UAV) photogrammetric point clouds because conventional data reduction suppresses filament-like structures. To address this persistent limitation, this study proposes a detection-guided pre-Structure from Motion (SfM) input-indexing workflow. You Only Look Once (YOLO) wire masks are inserted into image sets as class-indexed color while preserving Exchangeable Image File Format (EXIF) and exterior-orientation metadata for bundle adjustment. Following SfM-Multi View Stereo (MVS) densification, propagated class signals are extracted using elevation-referenced geometric filtering and a Red criterion exceeding 100 without voxel down-sampling. Matched reconstructions altered densified point counts by only 1.6% and overall processing time by 3.4%, indicating no added dense-matching penalty. Evaluated on 19 polylines totaling 423.1 m, wire-neighborhood points comprised 0.2584% of the geometrically filtered original cloud compared to 0.3191% in the class-indexed branch, representing a 23.5% gain. Segment-wise Random Sample Consensus (RANSAC) fitting across 415 one-meter segments confirmed improved spatial continuity: segment recovery increased from 56.1% to 67.7%, while maintaining an identical mean direction error of 1.60°. Additionally, the proposed branch finished faster at 8,331.1 s versus 8,622.9 s (-3.4%). Ultimately, detector-derived class information survives densification, significantly raising downstream vectorization yield.
No takes yet. Share an insight, caveat, or question.
Cho et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: