Reproduce “Automatic Failure Recovery and Re-Initialization for Online UAV Tracking with Joint Scale and Aspect Ratio Optimization” 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
Fangqiang Ding, Changhong Fu, Yiming Li, Jin Jin, Chen Feng
Researchers from Tongji University and New York University developed a real-time UAV tracking algorithm, JSAR, which dynamically adjusts both scale and aspect ratio, coupled with an efficient re-detection strategy (JSAR-Re) for long-term operations. The JSAR tracker achieved 35 frames per second on a CPU and outperformed many deep learning methods in precision and AUC on challenging UAV benchmarks, while JSAR-Re improved long-term tracking precision by 11.3% at 20 fps.
Key insights
- Joint optimization of scale and aspect ratio through a dedicated 2D size filter substantially improves tracking robustness, particularly against significant aspect ratio changes in UAV scenarios.
- By utilizing hand-crafted features within a DCF framework, high tracking performance can be achieved on a CPU, outperforming many deep learning trackers in specific UAV contexts.
- An efficient two-stage re-detection mechanism, combining CPU-friendly object proposal generation and a specialized decision filter, effectively enables long-term tracking with real-time performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.