SiamRPN with Multi-Scale Feature Fusion for Visual Object Tracking
Guangyan An · 2022 IEEE 5th International Conference on Automation, Electronics and Electrical Engineering (AUTEEE) · 2022
Visual object tracking aims to predict the bounding box in the whole sequence by giving an initial position, which has been paid more and more attention in recent years. The newly proposed Siamese trackers show great potential on high accuracy and robustness. In this paper we introduce a visual object tracker based on the Siamese region proposal network (SiamRPN) with multi-scale feature fusion. Specifically, it employs an additional feature fusion operation in the Siamese sub-network for feature extraction. It has achieved higher accuracy than the baseline SiamRPN without fine-tuning the original network and without considerable efficiency loss.