AUTHOR=Huang Dandan , Yu Siyu , Duan Jin , Wang Yingzhi , Yao Anni , Wang Yiwen , Xi Junhan 

TITLE=Spatio-temporal interactive fusion based visual object tracking method

JOURNAL=Frontiers in Physics

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1269638

DOI=10.3389/fphy.2023.1269638

ISSN=2296-424X

ABSTRACT=In visual object tracking tasks, the inter-frame correlation information of the object is not entirely utilized, and existing trackers are not well suited to solve problems such as local occlusion, deformations, and background interference.To address these issues, this paper proposes a visual object tracking method based on spatio-temporal interaction fusion. Firstly, by designing feature enhanced networks in both temporal and spatial dimensions, the spatio-temporal background information is fully utilized to obtain salient features that are more conducive to object recognition and improve the accuracy of object tracking. Moreover, it improves the adaptability of the model to object changes and reduces the drift of the model. Our proposed spatio-temporal interaction fusion network can additionally learn a similarity metric between the memory frame and the query frame based on feature enhancement. At the same time, it can filter out stronger feature representations through the interactive fusion of information from both. Finally, the tracking method proposed in this paper is validated on four challenging public datasets, and our method achieves SOTA with a substantial improvement in tracking performance in complex scenes affected by local occlusion, deformation, and background interference; it is also worth noting that the proposed method can achieve a success rate of 78.8% on the TrackingNet large scale dataset.