摘要
针对复杂跟踪环境条件下目标的跟踪失败问题,提出一种基于多相关滤波器组合的目标跟踪方法.首先2个分别采用颜色属性(Color Name,CN)特征和方向梯度直方图(Histogram of Oriented Gradient,HOG)特征的核相关滤波器(Kernelized Correlation Filter,KCF)通过自适应融合手段进行响应图信息融合,确定目标的预测位置;然后通过以目标区域为基础进行多尺度采样,提取CN-HOG拼接特征构建尺度相关滤波器,得到目标的最佳尺度;最后设计了模型的自适应更新策略,通过判断目标是否发生遮挡来决定是否在当前帧进行模型更新.在50组视频序列上对所提算法与6种当前主流的相关滤波跟踪算法进行了实验.实验结果表明,在复杂的跟踪环境条件下,所提算法取得了最好的跟踪精度和成功率,能够有效处理目标遮挡和尺度变化等问题,且具有较快的跟踪速度.
To cope with the problem of object tracking failure in the challenging environment, a target tracking method based on multi-correlation filter combination was proposed. Firstly, two kernelized correlation filters(KCF) based on color name(CN) features and histogram of oriented gradient(HOG) features, respectively, integrated the map information through adaptive fusion method, and were used to determine the prediction position of the target. Then, through the multi-scale sampling based on the target region, CN-HOG compositive feature was extracted to construct a scale correlation filter to obtain the optimal scale of target. Finally, the adaptive updating strategy of the model was designed to determine whether the model was updated in the current frame through determining whether the target was occluded. The proposed algorithm and 6 state-of-the-art methods were tested on 50 video sequences. The experiment results indicate that the proposed algorithm gains the best precision and success rate in the challenging environment, it can effectively deal with the problem of object occlusion and scale change, and it has a fast tracking speed.
作者
潘迪夫
李耀通
韩锟
PAN Difu;LI Yaotong;HAN Kun(School of Traffic and Transportation Engineering,Central South University,Changsha 410075,China)
出处
《湖南大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2019年第2期112-122,共11页
Journal of Hunan University:Natural Sciences
基金
国家自然科学基金资助项目(51305467)
湖南省自然科学基金资助项目(12JJ4050
2016JJ4117)~~
关键词
目标跟踪
相关滤波
尺度评估
模型自适应更新
object tracking
correlation filter
scale estimate
model adaptive updating