尺度方向自适应的相关滤波跟踪算法  

Correlated Filtering Tracking Algorithm with Adaptive Scale and Direction

作　　者：单玉刚 郝峰 SHAN Yu-gang;HAO Feng(Education Institute,Hubei University of Arts and Science,Xiangyang Hubei 441053,China)

机构地区：[1]湖北文理学院教育学院,湖北襄阳441053

出　　处：《计算机仿真》2021年第9期315-321,352,共8页Computer Simulation

基　　金：湖北省教育厅科学技术研究项目(B2016175);湖北文理学院博士基金(2015B002);湖北文理学院2019年度学科开放基金(XK2019069)。

摘　　要：针对DSST算法对目标方向发生变化时易出现的跟踪丢失问题,提出了一种目标尺度和方向自适应稳健跟踪算法。算法首先提取目标候选区域HOG和HSV特征,通过相关滤波算法构建多特征融合的二维定位滤波器,从而精确确定目标的中心位置。然后,根据方向池用HOG特征构建一维方向相关滤波器确定目标的最佳方向。并通过构建一维尺度相关滤波器确定最佳尺度。最后,根据PSR值变化情况调整相关滤波模型更新的权重,使模型适应目标的变化特征。选取OTB2013部分数据集进行测试,实验结果表明,上述算法距离精度保持在15pixels以内,成功率较DSST算法提高了20.1%,并且展示了上述算法对跟踪目标的尺度和方向变化具有鲁棒性和有效性。Aiming at the tracking loss of dsst algorithm when the target direction changes,a target scale and direction adaptive robust tracking algorithm is proposed.Firstly,the features of HOG and HSV in object candidate region were extracted,and a two-dimensional positioning filter based on multi-feature fusion was constructed by correlation filtering algorithm to determine the central position of the object accurately.Then,through multi-direction sampling of the object area,a one-dimensional directional correlation filter using HOG feature was constructed to determine the optimal direction of the object.then the optimal scale was determined by constructing a one-dimensional scale correlation filter.Finally,according to the change of PSR,the updating weight of correlation filter model was adjusted to make the model adapt to the change of the object features.Selecting partial data set of OTB2013 for test,the experimental results show that the distance accuracy of the proposed algorithm is kept within 15 pixels,and the success rate is 20.1%higher than that of DSST algorithm.It also shows that the proposed algorithm is robust and effective to the change of scale and direction of tracking object.

关 键 词：自适应 目标跟踪 相关滤波 尺度方向 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]