Siamese Network with Hybrid Fusion Strategy for RGB-T Tracking
Minjun Liang · 2023
Visual object tracking is to capture the target according to the initial target state, which can be widely used in autonomous driving, human interaction, smart surveillance. In the community of computer vision, the subject has become more popularized. With development of multi-modality vision, RGB-T tracking has been focused by researchers due to its robust performance and wider applications. Existing methods mainly focus on data fusion, which can be categorized into three manners, including image fusion, feature fusion and decision fusion. In this paper, we aim to analyze the complementary role of various fusion types. We proposed a hybrid fusion strategy, which combines those types into a framework. We select Siamese based tracker as baseline and evaluate them in GTOT dataset. Our fusion strategy obtains the best performance against all other fusion modules, which achieves 65.3% and 77.8% success rate and precision rate. Our method can validate the complementary role of different fusion methods.