摘要
鲁棒的尺度判别一直是视频目标跟踪领域中一个富有挑战性的问题.现有的算法在处理复杂图像序列的尺度变化问题时,跟踪速度和精度方面都还有待提升.本文构建两个相关滤波器,加入尺度变换,对目标跟踪的尺度参数进行优化,以提升跟踪速度和精度.首先,构建一维和二维相关滤波器,其中二维位置滤波器实现目标的跟踪以确定目标的位置,一维尺度滤波器对尺度变换进行初步计算得到目标的尺度.然后,组合一维和二维相关滤波器形成三维滤波器,实现最终的目标定位;最后,分析尺度因子参数的取值对跟踪中的速度、中心位置偏移、位置精度和重叠精度的影响.在OTB-2015数据集进行实验,获得了目标跟踪尺度参数的优化取值.
Robust scale estimation has been a challenging problem in target tracking. In handling complex scale variation of the image sequence,existing algorithms have yet to be promoted in tracking speed and tracking precision. We build two related filters and introduce the scale transformation,and optimize the scale parameters of the target tracking,which can enhance the tracking speed and precision. First,this paper constructs a 1-dimensional correlation filter and a 2-dimensional correlation filter,and the 2-dimensional filter realizes the target tracking,determines the location of the object. The evaluation of scale transformation is realized by 1-dimensional filter. Then,the two filters are combined into a 3-dimensional filter to complete the detailed dimension space target positioning. Finally,we analyze the effect of scale factor on the tracking speed,centre location error,distance precision and overlap precision to obtain the optimized value. Experiments are performed on the data set OTB-2015,and the optimized value of target tracking scale parameters are acquired.
作者
吴慧君
李梅云
杨文元
Wu Huijun;Li Meiyun;Yang Wenyuan(School of Information Engineering,Zhangzhou Institute of Technology,Zhangzhou 363000,China;Fujian Key Laboratory of Granular Computing and Application,Minnan Normal University,Zhangzhou 363000,China)
出处
《南京师大学报（自然科学版）》
CAS
CSCD
北大核心
2019年第4期69-76,共8页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学青年基金项目(61703196)
福建省自然科学基金项目(2018J01549)
关键词
计算机视觉
目标跟踪
相关滤波
尺度估计
参数优化
computer vision
target tracking
correlation filtering
scale estimation
parameter optimization
