摘要
为解决单一特征目标跟踪算法因光照变化、目标遮挡导致的跟踪失败问题,提出一种自适应加权融合颜色属性特征和方向梯度直方图特征的核相关滤波目标跟踪算法.首先根据目标颜色属性特征和方向梯度直方图特征,采用核相关滤波方法获得2种特征下目标的预测位置;然后分别计算2个目标位置的核相关滤波响应值,据此响应值大小按比例分配目标位置权重,获得目标的最终预测位置,提高了目标定位的精度;最后采用相邻2帧图像的平均差来分析图像的变化率,按变化率分段调整分类器的学习速率,解决目标遮挡导致的跟踪失败问题.在9组标准测试视频集下验证文中算法,并与3种核相关滤波跟踪算法进行比较的结果表明,与其中最优算法相比,该算法跟踪目标的平均中心位置误差减少16.78个像素,平均距离精度提高11.01%,平均重叠精度提高14.87%;在目标严重遮挡情况下,该算法依然能够稳定地跟踪目标.
In order to solve the problem of object tracking failure caused by object occlusion in object trackingalgorithm based on a single feature,this paper proposes an object tracking algorithm based on adaptivefeature fusion of color name(CN)feature and histogram of oriented gradient(HOG)feature.Firstly,a kernelizedcorrelation filter method based on CN features and HOG features respectively is applied to predictthe positions of the object.Then the final object location is obtained by assigning the filter response valuesas their corresponding weights for the two predicted positions,so that the accuracy of object location is improved.Next,the change rate of image frames by mean error of two adjacent frames is used to piece-wiselyadjust the learning rate of the classifier.This strategy solves the problem of tracking failure caused by objectocclusion.Finally,9groups of standard test video sequences are tested to verify the performance of theproposed algorithm.Compared with the best among the other three correlation filtering tracking algorithms,the average center location error is reduced by16.78pixels,the average distance precision is increased by11.01%and the average overlapping precision is increased by14.87%.The proposed tracking method is stillable to track the moving object stably and accurately in the case of severe occlusion of the object.
作者
熊昌镇
赵璐璐
郭芬红
Xiong Changzhen;Zhao Lulu;Guo Fenhong(Beijing Key Laboratory of Urban Intelligent Control, Beijing 100144;School of Sciences, North China University of Technology, Beijing 100144)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2017年第6期1068-1074,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
北京市属高等学校青年拔尖人才培育计划(CIT&TCD201404009)
科技创新服务能力建设-科技成果转化-提升计划项目(PXM2016_014212_000036)
关键词
目标跟踪
核相关滤波
目标遮挡
特征融合
object tracking
kernelized correlation filter
object occlusion
feature fusion
