Improved Twin Network Target Tracking Algorithm Based on Hybrid Attention Mechanism

Abstract

Deep getting to know techniques are turning into extra and extra famous in the area of goal monitoring due to their effective function extraction capabilities, amongst which the twin network-based goal monitoring algorithm with excessive walking body price and robust monitoring fault tolerance has attracted the interest of many scholars. Aiming at the drawbacks that twin networks use an offline method to recognize end-to-end mannequin training, and the educated mannequin can't trade the community parameters and the goal template at some point of the monitoring process, this paper proposes an increased twin community goal monitoring algorithm primarily based on a hybrid interest mechanism, which is first off skilled offline on the education set of the ILSVRC2015 dataset, and then on the coaching units of ILSVRC2015 and The algorithm is examined and evaluated on a take a look at set on the plane class of the VOT2018 dataset, and the experimental outcomes exhibit that the success fee of the algorithm proposed in this paper is multiplied from 61.2% to 68.4% and the accuracy from 77.1% to 82.3% of the SiamFC base algorithm, which proves that the algorithm has properly monitoring performance.