Reproduce “Vision-Based Power Line Cables and Pylons Detection for Low Flying Aircraft” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jakub Gwizdała, Doruk Oner, Soumava Kumar Roy, Mian Akbar Shah, Ad Eberhard, Ivan Egorov, Philipp Krüsi, Grigory Yakushev and 1 more

Researchers from EPFL and Daedalean AG developed DDLN, a deep learning system for jointly detecting power line cables and pylons from aircraft-mounted cameras at long distances. The system demonstrated improved detection performance over baselines, achieving a Quality score of 0.72 for cables on a challenging dataset, and operates in real-time (5 FPS) on onboard hardware for enhanced flight safety.

Key insights

  • Pre-trained weights from large datasets like ImageNet are essential for training robust detection models for distant, barely visible aerial objects.
  • A composite loss function incorporating a connectivity term (L_MALIS) is vital for accurately detecting curvilinear structures like power cables, preventing topological errors.
  • Jointly detecting both power lines and pylons can be achieved without compromising individual cable detection performance, possibly benefiting from contextual information.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.