摘要
针对目标遮挡导致跟踪失败的问题,在核相关滤波器(KCF)的基础上,提出目标重检测的长期跟踪算法.将方向梯度直方图特征与亮度颜色空间(LAB)颜色信息融合,建立目标外观模型,并增加尺度估计对应目标尺度的改变.通过引入峰值比控制重检测模块的启动,提取Harris角点重新学习新的模型,对目标进行持续跟踪,并控制模型更新速率.在OTB数据集上的对比实验表明,文中算法跟踪精度较高,适用于遮挡情况下的目标长期跟踪.
Based on the kernelized correlation filter(KCF) algorithm, a long-term tracking algorithm combining target re-detection is proposed. Firstly, the histogram of oriented gradient features and LAB color information are fused, the appearance model is established and scale estimation is added to cope with the changes of object scale. Then, the peak ratio is introduced to control the start of re-detection module and a correlation filter model is re-learned by extracting Harris corners. Finally, the occluded object is continuously tracked with the proposed model updating strategy. Comparison experiments on OTB datasets show that the proposed algorithm produces higher tracking accuracy and is suitable for long- term tracking with occlusion.
作者
李娜
吴玲风
李大湘
LI Na1;2;3;WU Lingfeng1;2;3 LI Daxiang1;2;3
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2018年第10期899-908,共10页
Pattern Recognition and Artificial Intelligence
基金
陕西省国际合作交流项目(No.2017KW-013)
陕西省教育厅专项科研计划项目(No.17JK0692)
西安邮电大学研究生创新基金项目(No.CXJJ2017004)资助~~