结合多注意力机制的Transformer实时目标跟踪算法  

Transformer Real-Time Target Tracking Algorithm Combining Multiple Attention Mechanisms

作　　者：黄易仟 徐杨 张永丹 肖慈 冯明文 HUANG Yiqian;XU Yang;ZHANG Yongdan;XIAO Ci;FEMG Mingwen(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;Guiyang Aluminum-Magnesium Design and Research Institute Co.Ltd.,Guiyang 550009,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵阳550025 [2]贵阳铝镁设计研究院有限公司,贵阳550009

出　　处：《计算机工程与应用》2024年第23期187-197,共11页Computer Engineering and Applications

基　　金：贵州省科技计划项目(黔科合支撑[2023]一般326)。

摘　　要：近年来目标跟踪在学术研究和实际应用中具有巨大潜力,受到越来越多的关注并成为计算机视觉的热点研究方向。针对现有的目标跟踪算法在光照变化、遮挡和快速运动复杂背景条件下跟踪精度较低这一问题,提出一种结合多注意力机制的Transformer跟踪算法——TrKYS。引入多注意力模块捕捉目标在空间和通道维度中的特征,实现对目标特征上下文依赖关系的建模;通过多个不同空洞率的平行空洞卷积对特征图进行采样,以获得图像的多尺度特征,增强局部特征表达能力;构建Exemplar Transformer模块,利用模板化的空间信息和目标特征建立了外观跟踪模型。为了适应外观变化,该跟踪算法在连续帧中跟踪目标物体时会实时更新跟踪模板特征向量、运动参数和空间信息,提高跟踪定位的精度。在LaSOT、VOT2018、NFS、OTB-2015、TrackingNet和GOT-10k数据集上实验结果表明,与其他主流目标跟踪算法相比,所提算法具有更好的跟踪性能,特别是在TrackingNet上的跟踪算法(TrKYS)相对于基准跟踪算法(KYS)成功率提高了3个百分点。In recent years,target tracking has great potential in academic research and practical application,and has received more and more attention and become a hot research direction of computer vision.Aiming at the problem that the existing target tracking algorithms have low tracking accuracy under complex background conditions of illumination change,occlusion and fast motion,this paper proposes a Transformer tracking algorithm combined with multi-attention mechanism-TrKYS.Firstly,the multi-attention module is introduced to capture target features in spatial and channel dimensions,and the modeling of context dependence for target feature is realized.Then,the feature map is sampled by multiple parallel hole convolution with different void rates to obtain multi-scale features of the image and enhance local feature expression ability.Finally,the Exemplar Transformer module is constructed,and the appearance tracking model is established by using templated spatial information and target features.In order to adapt to the appearance change,the tracker will update the tracking template feature vector,motion parameters and spatial information in real time when tracking the target object in consecutive frames,improving the accuracy of tracking and positioning.Experimental results on LaSOT,VOT2018,NFS,OTB-2015,TrackingNet and GOT-10k datasets show that in comparison with other mainstream target tracking algo-rithms,the proposed algorithm has better tracking performance,in particular,the tracking algorithm in this paper(TrKYS)on TrackingNet improves the success rate by 3 percentage points relative to the benchmark tracking algorithm(KYS).

关 键 词：Exemplar Transformer 目标跟踪 特征增强 外观变化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]