Single-objective Tracking Algorithm Based on Siamese Networks

Zong-cheng Miao,Shi-yan Gao,Ze-min He,Yuan Ou
DOI: https://doi.org/10.37188/cjlcd.2022-0186
2023-01-01
Chinese Journal of Liquid Crystals and Displays
Abstract:In the computer vision applications, the tracking algorithms based on the Siamese networks have improved in speed and accuracy compared with the traditional target tracking algorithms, but they are greatly affected by interference factors such as occlusion and deformation. Based on these, the existing target tracking methods and technologies based on Siamese networks are summarized and analyzed, they mainly include the introduction of fully convolutional Siamese neural network method, regression method and online update method in Siamese networks, and the improvements of the target tracking algorithms based on three methods are reviewed. The research progress and development status of Siamese networks in target tracking applications in recent years are introduced in detail. Then, the VOT2017 and LaSOT datasets are used for experimental comparison, and the performances of various tracking algorithms based on Siamese neural networks are compared. At the end, the development trend of the target tracking methods based on Siamese neural networks is prospected.
What problem does this paper attempt to address?