Double stage Siamese network object tracking algorithm based on template update

Aiming at the problem that the tracking accuracy of SiamRPN declines when the appearance of the object changes dramatically, this paper proposes a double stage Siamese network object tracking algorithm based on template update. The algorithm can be divided into two stages. In the first stage, the SiamRPN was used for preliminary object tracking to get the initial prediction, updating network was introduced to update the object template in the second stage. Pixel-wise correlation and corner prediction head were applied in this algorithm to improve the feature representation. A large number of experiments on OTB2015 and VOT2018 show that our approach has strong robustness for the appearance of the object dramatic change, and was able to improve tracking results of comparison algorithm, effectively improves the tracking precision and success rate.