Deep Snake协助相关滤波目标跟踪  

Deep Snake-Assisted Correlation Filters for Object Tracking

作　　者：李豪 LI Hao(Unit 75220,Huizhou 516133,China)

机构地区：[1]75220部队,广东惠州516133

出　　处：《信息工程大学学报》2024年第5期538-544,共7页Journal of Information Engineering University

摘　　要：针对在相关滤波目标跟踪中,对目标尺度观测不准会造成匹配模板更新错误,从而降低目标跟踪的精度的问题,提出一种Deep Snake协助相关滤波的目标跟踪方法。该方法首先利用相关滤波估计目标的初始状态;然后利用Deep Snake得到目标的轮廓,进而优化目标的尺度。在OTB-2015(成功率70%)和VOT-2019(期望平均覆盖率28.6%)数据集上的实验结果表明:与现有的目标跟踪方法相比,提出的跟踪方法具有较优的跟踪性能。To address the problem that inaccurate observation of target scale will result in incorrect up⁃dating of matching template and reduce the accuracy of target tracking in correlation filtering,an ob⁃ject tracking method assisted by deep snake with correlation filters is proposed.The method consists of two stages.In the first stage,the initial state of target is estimated by the correlation filter.In the sec⁃ond stage,deep snake is used to get the outline of the target and then optimize its scale.Experimental results on OTB-2015(70%Success)and VOT-2019(28.6%EAO)datasets show that the proposed method has better tracking performance compared with the state-of-the-art methods.

关 键 词：目标跟踪 相关滤波 主动轮廓模型 目标尺度观测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]