Abstract
• Novel RGB-T tracking algorithm based on modality difference reduction and effective information re-selection. • Selectively promoting cross-modality feature interactions for modality difference reduction. • Adaptively selecting discriminative multi-modal and unimodal features for RGB-T tracking. • State-of-the-art performance on most popular RGB-T tracking benchmarks with real-time tracking speed. RGB-T tracking has attracted increasing attention, since visible and thermal data have strong complementary advantages to improve the robustness of trackers. Most existing models focus on investigating efficient ways of fusing the complementary information from RGB and thermal images for better tracking performance. However, the modality differences caused by different imaging mechanisms may degrade the discriminability of the fused features. Meanwhile, compared with the unimodal features, the fused features may not always improve the tracking performance, especially when one of the input images contain much noisy information. In view of this, we propose a novel RGB-T tracking model by simultaneously reducing modality difference and re-selecting discriminative features from the fused features as well as from the unimodal features. To this end, a Modality Difference Compensation module (MDC) and a Feature Re-selection module (FRS) are presented. The former one reduces the modality differences between RGB and thermal features and obtains the fused features. The latter one adaptively selects such discriminative features from the unimodal features and the fused features for the subsequent classification and regression. Exhausted experiments are conducted on three RGB-T tracking benchmark datasets, which verify that our proposed tracker performs favorably against some state-of-the-art tracking algorithms.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.