复杂场景下一种改进的单目标跟踪算法研究
An Improved Single Target Tracking Algorithm for Complex Scenes
侯艳丽 1魏义仑 1王鑫涛 1王娟1
作者信息
- 1. 河北科技大学信息科学与工程学院,河北 石家庄 050018
- 折叠
摘要
针对部分复杂场景下目标跟踪存在跟踪框漂移问题,基于孪生候选区域生成网络(SiamRPN),融合通道注意力模块和选择核心模块(SK Module),提出一种单目标跟踪算法CAKSiamRPN.特征提取部分引入高效通道注意模块(ECAM)和基于标准化的通道注意力模块(NCAM),在不降低通道维度的情况下,摒弃相似信息,突出显著特征,关注并提取特定信息;在候选区域生成网络(RPN)嵌入SK Module,增强全局信息嵌入,减少填充操作的影响.将改进算法与SiamRPN及其它经典单目标跟踪算法在OTB100 和UAV123 数据集上进行实验对比.实验结果表明:跟踪精度和成功率明显提高,能更好地适应复杂场景,具有更强的鲁棒性.
Abstract
To address the problem of tracking frame drift for target tracking in some complex scenarios,based on the Siamese region proposal network(SiamRPN),incorporating the channel attention module and the selection core module(SK Module),a single-target tracking algorithm CAKSiamRPN is proposed.The feature extraction part intro-duces efficient channel attention module(ECAM)and normalization-based channel attention module(NCAM)to dis-card similar information,highlight salient features,and focus on and extract specific information without reducing the channel dimensionality;the SK Module is embedded in the Region Proposal Network(RPN)to enhance global infor-mation embedding and reduce the impact of padding operations.The improved algorithm was compared with SiamRPN and other classical single-target tracking algorithms on the OTB100 and UAV123 datasets.The experimental results show that the tracking precision and success rate are significantly improved,and it can better adapt to complex scenes with stronger robustness.
关键词
目标跟踪/复杂场景/孪生网络/注意力模块/选择核心模块Key words
Target tracking/Complex scenes/Siamese network/Attention module/Selective kernel module引用本文复制引用
出版年
2024
NETL
NSTL
万方数据