Reproduce “DeTracker: Motion-decoupled Vehicle Detection and Tracking in Unstabilized Satellite Videos” 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
Jiajun Chen, Jing Xiao, Shaohan Cao, Yuming Zhu, Liang Liao, Jun Pan, Mi Wang
Researchers from Wuhan University developed DeTracker, a framework for motion-decoupled vehicle detection and tracking in unstabilized satellite videos, circumventing the need for prior stabilization. The method achieved an average MOTA improvement of 5.1% and IDF1 improvement of 2.9% on the new SDM-Car-SU dataset, demonstrating enhanced performance on real unstabilized satellite footage.
Key insights
- Explicitly decoupling global background motion from local target motion directly in the feature space allows for more robust motion correction than traditional image-level stabilization.
- Aggregating temporal dependencies and fusing multi-scale features across consecutive frames significantly enhances the continuity and discriminability of tiny-object representations in unstable videos.
- A new simulated dataset, SDM-Car-SU, systematically quantifies algorithm robustness against various platform motion perturbations, filling a gap for unstabilized satellite video research.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.