MDBST: Multi-Different-Branch Siamese Network for Object Tracking
Hao Zhang, Yan Piao, Bailiang Huang · 2022 IEEE 2nd International Conference on Power, Electronics and Computer Applications (ICPECA) · 2022
In the field of machine vision, object tracking tasks have always been a hot issue for many researchers. In recent years, many excellent trackers based on deep learning have achieved very advanced performance in object tracking tasks. However, many trackers use a single-branch feature extraction network to design trackers. Researchers are constantly increasing the depth of the feature extraction network while ignoring the width of the network. If the depth of the network is too deep, the size of the output feature map will be too small, which will affect the tracking results. Therefore, in order to balance the depth and feature extraction ability of the feature extraction network, this paper proposes a multi-different-branch feature extraction network. We design a feature extraction module and add the residual module at the same time, design two different branches to perform feature extraction in parallel, and concatenate the feature maps output by the two branches. Effectively extract all aspects of feature information of the object, and improve the accuracy and robustness of target tracking. We embed the feature extraction network into siamese network and conduct extensive experiments on the OTB2015 and VOT2016 benchmark datasets. The results show that our MDBST tracker has achieved excellent results.