Real-time RGBT Tracking Algorithm Based on Multi-modal Interactive Fusion
Qian Zhao, Xingzhong Xiong, Junjia Wang, Jun Liu · 2024
A real-time RGBT tracking method based on multimodal interaction fusion is suggested to address the issue that multi-domain network-based RGBT target tracking algorithms find it challenging to concurrently consider robustness and realtime performance. It is designed with an information interaction fusion module that can use multi-level features to get more information. Global information from both channel and space dimensions is then fused using a channel space fusion module, which effectively fuses two modal information to enhance tracking performance. The algorithm in this paper is experimentally validated on the GTOT, RGBT234 and LasHeR datasets. The results demonstrate that the algorithm can guarantee robustness while achieving real-time performance, with PR/SR reaching 90.5%/73.2%, 84.9%/61.8%, and 48.1%/34.3%, respectively, and a running speed of 41.07 fps.