融合注意力机制的单步双分支多目标跟踪算法  

Single-step dual-branch multi-object tracking algorithm fused with attention mechanism

作　　者：张红艳 黄宏博 何嘉玉[1] 刘亚辉 李颖[1] ZHANG Hong-yan;HUANG Hong-bo;HE Jia-yu;LIU Ya-hu;LI Ying(School of Computer,Beijing Information Science and Technology University,Beijing 100101,China;Institute of Computational Intelligence,Beijing Information Science and Technology University,Beijing 100192,China;School of Information Management,Beijing Information Science and Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学计算机学院,北京100101 [2]北京信息科技大学计算智能研究所,北京100192 [3]北京信息科技大学信息管理学院,北京100192

出　　处：《计算机工程与设计》2022年第11期3085-3092,共8页Computer Engineering and Design

基　　金：北京市教委科技计划一般基金项目(KM201811232024);北京信息科技大学科研基金项目(2021XJJ30、2021XJJ34)。

摘　　要：为提高多目标跟踪的实时性、解决形变和遮挡后的身份切换问题,提出一种融合改进的通道注意力机制的单步双分支多目标跟踪算法。以深度聚合网络为框架,通过改进通道注意力机制,充分利用多层次特征并考虑特征通道依赖关系;联合训练目标检测分支与重识别分支,增强检测鲁棒性和模型抗遮挡能力;利用外观、运动以及IOU信息进行数据关联。实验结果表明,使用该算法有效缓解了多目标跟踪遇到的目标形变和遮挡问题带来的影响,性能和效率得到大幅提升。To improve the real-time performance of multi-object tracking and solve the problem of identity switching after defor-mation and occlusion,a single-step dual-branch multi-object tracking algorithm incorporating an improved channel attention mechanism was proposed.Taking the deep aggregation network as the framework,by improving the channel attention mechanism,multi-level features were utilized and the dependency of feature channels was considered.The target detection branch and re-identification branch were jointly trained to enhance the detection robustness and the model’s anti-occlusion ability.The appearance,movement,and IOU information were used for data association.Experimental results show that the effects of target deformation and occlusion problems encountered in multi-object tracking are effectively alleviated,and the performance and efficiency are greatly improved.

关 键 词：注意力机制 多目标跟踪 深度聚合 卡尔曼滤波 匈牙利算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]