SiamOAN: Siamese object-aware network for real-time target tracking
文献类型:期刊论文
作者 | Wei BB(魏冰冰)1,2,3,4; Chen HY(陈宏宇)2,3,4; Ding QH(丁庆海)5; Luo HB(罗海波)2,3,4 |
刊名 | Neurocomputing |
出版日期 | 2022 |
卷号 | 471页码:161-174 |
ISSN号 | 0925-2312 |
关键词 | Siamese-based tracking Global context Background interference Object tracking |
产权排序 | 1 |
英文摘要 | Existing Siamese-based tracking algorithms usually utilize local features to represent the object, which lack sufficient discrimination and may degrade tracking performance in challenging situations. To address this issue, we propose a novel object-aware network to improve feature representation and achieve robust object tracking. The proposed object-aware network contains a background filter module (BFM), channel complementary module (CCM), and template adaptive network (TAN). Specifically, by locating the target in the initial frame on the feature maps, BFM suppresses the background interference of the target template. CCM captures the global context by exploring the complementary information of each channel. The lightweight TAN adaptively recognizes valuable features for the target and represents the target template just through a single vector. Benefiting from these three components, the object-aware network enhances the discrimination of feature maps and alleviates background interference to some extent. The proposed object-aware network could be integrated with the Siamese-based backbone network for real-time object tracking, named SiamOAN. Extensive experiments on the six challenging benchmarks including OTB100, UAV123, VOT2016, VOT2018, GOT10k, and LaSOT, show that the proposed SiamOAN outperforms many state-of-the-art trackers and runs at approximately 67 fps on GPU RTX3090. |
WOS关键词 | ROBUST |
WOS研究方向 | Computer Science |
语种 | 英语 |
WOS记录号 | WOS:000761907400006 |
源URL | [http://ir.sia.cn/handle/173321/30087] |
专题 | 沈阳自动化研究所_光电信息技术研究室 |
通讯作者 | Wei BB(魏冰冰) |
作者单位 | 1.University of Chinese Academy of Sciences, Beijing 100049, China 2.Key Laboratory of Opto-Electronic Information Processing, Chinese Academy of Sciences, Shenyang 110016, China 3.Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China 4.Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110169, China 5.Space Star Technology Co, Ltd., Beijing 100086, China |
推荐引用方式 GB/T 7714 | Wei BB,Chen HY,Ding QH,et al. SiamOAN: Siamese object-aware network for real-time target tracking[J]. Neurocomputing,2022,471:161-174. |
APA | Wei BB,Chen HY,Ding QH,&Luo HB.(2022).SiamOAN: Siamese object-aware network for real-time target tracking.Neurocomputing,471,161-174. |
MLA | Wei BB,et al."SiamOAN: Siamese object-aware network for real-time target tracking".Neurocomputing 471(2022):161-174. |
入库方式: OAI收割
来源:沈阳自动化研究所
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。