尺度自适应相关滤波跟踪算法  

Scale Adaptive Correlated Filter Tracking Algorithm

作　　者：王一博 郭培胜[2] 陆晓燕 汪波 李新德[1] WANG Yibo;GUO Peisheng;LU Xiaoyan;WANG Bo;LI Xinde(School of Automation,Southeast University,Nanjing 210096,China;The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210007,China)

机构地区：[1]东南大学自动化学院,南京210096 [2]中国电子科技集团公司第二十八研究所,南京210007

出　　处：《指挥信息系统与技术》2021年第3期35-41,共7页Command Information System and Technology

基　　金：国家自然科学基金(62073072);信息系统工程重点实验室开放基金(05201905)资助项目。

摘　　要：针对相关滤波跟踪器难以处理目标尺寸变化问题,提出了一种基于特征融合和时空正则化的尺度自适应相关滤波跟踪(SACFT)算法,同时结合方向梯度直方图(HOG)特征和颜色特征,以进一步提高整体跟踪性能。使用时空正则化相关滤波器处理边界效应问题,并用交替方向乘子法(ADMM)求解目标函数,使其收敛到全局最优。最后,不同场景的定量和定性试验结果表明该算法在目标尺度变化方面具有较好的鲁棒性。Aimed at the problem of the correlated filter tracker about difficult to deal with the change of the target’s size,a scale adaptive correlation filter tracking(SACFT)algorithm based on feature fusion and spatio-temporal regularization is proposed.Meanwhile,the histogram of oriented gradient(HOG)features and the color features are combined to further enhance the overall tracking performance.The spatio-temporal regularized correlation filter is used to deal with the boundary effect problem,and the alternating direction method of multipliers(ADMM)is adapted to solve the objective function.Thus,the function can converge to the global optimization.Finally,the quantitative and qualitative experimental results in different scenes show that the algorithm has good robustness on the change of the target’s size.

关 键 词：视觉跟踪 相关滤波 时空正则化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]