
Target tracking algorithm of intermittently updated template based on reliability evaluation
Author(s) -
Qingsong Xie,
Jingwei Shen,
Zhiyong An
Publication year - 2022
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2216/1/012073
Subject(s) - tracking (education) , computer science , template matching , template , artificial intelligence , tracking system , similarity (geometry) , computer vision , reliability (semiconductor) , key (lock) , relevance (law) , frame (networking) , algorithm , kalman filter , image (mathematics) , psychology , telecommunications , pedagogy , power (physics) , physics , computer security , quantum mechanics , political science , law , programming language
In the siamese network tracking method based on the target matching idea, the tracking performance is closely related to the similarity between the tracking template and the tracking target. The UpdateNet algorithm updates the tracking results of each frame prediction into the template during tracking, which undoubtedly guarantees the real-time correlation between the template and the tracking target. However, tracking targets will encounter tracking challenges such as occlusion and short-term severe deformation. The tracking results predicted under this tracking challenge will be mixed with messy information that has nothing to do with the tracking target. If the UpdateNet algorithm updates the tracking results that are rich in messy information into the template, the purity of the template will decrease. This paper proposes a method of intermittently updating templates by proposing template update constraints, so that the UpdateNet algorithm discards the low-confidence tracking results. This guarantees the purity of the tracking template to a certain extent, which improves the relevance of the tracking template and the tracking target. In addition, this article also conducts optimal training on the UpdateNet algorithm.