EUVIMEDEUVIMED entdecken
Crossref

GPTrack: Graph-Topological Priors and Physics-Aware Interaction for RGB-T Tracking

Jingjing Li, Hanlin Qin, Yue Yu, Xupei Zhang, Juliu Li

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 17362-17377

DOI: 10.1109/tcsvt.2026.3719072 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Jingjing Li und 4 weiteren Autor:innen. Die Arbeit nennt 72 Literaturangaben. Gefördert von: Fundamental Research Funds for the Central Universities, Postgraduate Innovation Fund of Xidian University, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Current RGB-T tracking methods exploit the complementary characteristics of visible and thermal sensing; however, most still rely on late-stage fusion and purely data-driven attention. These architectures have two primary limitations: delayed interaction amplifies single-sensor degradations across deep layers, and similarity-based fusion often ignores the modality-specific imaging characteristics of heterogeneous optoelectronic sensors. To address these issues, we propose GPTrack, a unified framework driven by graph-topological priors and physics-aware interaction. Specifically, we introduce a Multi-Relation Topology Graph (MRTG) before backbone encoding to establish early spatial, cross-modal, and semantic connections, thereby preserving fine-grained structures and alleviating degradation accumulation across deep layers. Furthermore, we propose a Geometry-Saliency Cross-Attention (GSCA) module to integrate imaging-inspired physical guidance into cross-modal interaction. By combining visible spatial gradients with infrared thermal saliency, GSCA encodes structural variations and target-aware thermal cues into a joint guidance mask for attention-based feature aggregation. Modulated by an adaptive residual gating, this mask regulates feature interaction and suppresses background interference. Extensive experiments on five public benchmarks demonstrate that GPTrack achieves state-of-the-art performance across diverse scenarios. The code is publicly available at https://github.com/qixi-alt/GPTrack

Förderung

  • Fundamental Research Funds for the Central Universities
  • Postgraduate Innovation Fund of Xidian University
  • National Natural Science Foundation of China
  • Shaanxi Province Key Research and Development Plan Project
  • Shaanxi Qinchuangyuan “Scientist+Engineer” Team Construction Project

Zitieren

APA

Li, J., Qin, H., Yu, Y., Zhang, X., & Li, J. (2026). GPTrack: Graph-Topological Priors and Physics-Aware Interaction for RGB-T Tracking. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 17362-17377. https://doi.org/10.1109/tcsvt.2026.3719072

Vancouver

Li J, Qin H, Yu Y, Zhang X, Li J. GPTrack: Graph-Topological Priors and Physics-Aware Interaction for RGB-T Tracking. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):17362-17377. doi:10.1109/tcsvt.2026.3719072

BibTeX
@article{Li2026,
  author = {Li, Jingjing and Qin, Hanlin and Yu, Yue and Zhang, Xupei and Li, Juliu},
  title = {GPTrack: Graph-Topological Priors and Physics-Aware Interaction for RGB-T Tracking},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {17362--17377},
  doi = {10.1109/tcsvt.2026.3719072}
}
Kostenlos starten →