Open AccessSiamese adversarial network for object tracking
Author(s) -
Kim H.I.,
Park R.H.
Publication year - 2019
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2018.7104
Subject(s) - discriminator , bittorrent tracker , artificial intelligence , computer science , adversarial system , object (grammar) , similarity (geometry) , computer vision , video tracking , generative adversarial network , residual , tracking (education) , deep learning , image (mathematics) , algorithm , eye tracking , telecommunications , psychology , pedagogy , detector
In this Letter, a Siamese adversarial network tracker (SANT) is proposed. Recently, in computer vision field, generative adversarial network (GAN) has been widely used for image and video generation. Using the GAN, the proposed method constructs a Siamese adversarial network (SAN) for object tracking. Unlike existing GANs, the proposed SAN uses similarity learning with SAN discriminator. To show the effectiveness of the proposed SAN, the same structure as the residual long‐short‐term memory tracker is used. Experimental results show that the proposed SANT achieves the highest performance among existing Siamese trackers.