EUVIMEDScopri EUVIMED
Crossref

Frequency-Guided and Temporal Prompt Information Integration for RGB-T Tracking

Yunzhen Li, Bineng Zhong, Qihua Liang, Yaozong Zheng, Ning Li, Shuxiang Song

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 17843-17854

DOI: 10.1109/tcsvt.2026.3718653 ↗

At a glance

Journal article in IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Written by Yunzhen Li and 5 co-authors. 65 references. Funded by: Project of Guangxi Science and Technology, National Natural Science Foundation of China, Guangxi “Young Bagui Scholar” Teams for Innovation and Research Project.

AbstractDati bibliograficiCiteRelated studies

Abstract (original language)

Frequency characteristics constitute an important distinction between visible and infrared images in RGB-Thermal (RGB-T) tracking. However, existing RGB-T trackers emphasize token-level interaction while ignoring the frequency-domain differences essential to their complementarity, which constrains effective cross-modal interaction. Furthermore, these trackers focus on intra-frame cross-modal interaction but overlook temporal dependencies across frames, resulting in limited robustness under challenging scenarios. To overcome these limitations, we propose FITrack, a novel RGB-T tracking framework that converts modality-specific frequency complementarity into temporal-state-aware attention guidance. The proposed architecture consists of two key components: a Temporal Prompt Modulation (TPM) module and a Frequency-guided Attention Modulation (FAM) module. The TPM summarizes key tokens from previous frames into compact temporal prompts, which condition adaptive frequency-mask generation and preserve temporal continuity during tracking. Additionally, the FAM module performs Fourier transform-based decomposition to select modality-preferred spectral components and injects TIR low-frequency structural cues into RGB high-frequency attention logits. By coupling TPM and FAM, FITrack performs temporal prompt-conditioned frequency selection and frequency-guided attention modulation, producing discriminative and temporally stable target representations. Extensive experiments on three widely used RGB-T benchmarks, including RGBT210, RGBT234, and LasHeR, demonstrate that FITrack achieves competitive performance against recent RGB-T and RGB-X trackers. The results validate the effectiveness of temporal prompt-conditioned frequency selection and frequency-guided attention modulation for robust RGB-T tracking in diverse and challenging scenarios.

Funding

  • Project of Guangxi Science and Technology
  • National Natural Science Foundation of China
  • Guangxi “Young Bagui Scholar” Teams for Innovation and Research Project
  • Research Project of Guangxi Normal University
  • Guangxi Engineering Research Center of Educational Intelligent Technology

Cite

APA

Li, Y., Zhong, B., Liang, Q., Zheng, Y., Li, N., & Song, S. (2026). Frequency-Guided and Temporal Prompt Information Integration for RGB-T Tracking. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 17843-17854. https://doi.org/10.1109/tcsvt.2026.3718653

Vancouver

Li Y, Zhong B, Liang Q, Zheng Y, Li N, Song S. Frequency-Guided and Temporal Prompt Information Integration for RGB-T Tracking. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):17843-17854. doi:10.1109/tcsvt.2026.3718653

BibTeX
@article{Li2026,
  author = {Li, Yunzhen and Zhong, Bineng and Liang, Qihua and Zheng, Yaozong and Li, Ning and Song, Shuxiang},
  title = {Frequency-Guided and Temporal Prompt Information Integration for RGB-T Tracking},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {17843--17854},
  doi = {10.1109/tcsvt.2026.3718653}
}
Start for free →