基于特定目标提议框的自适应跟踪算法  

Adaptive Tracking Algorithm Based on Specific Object Proposal

作　　者：胡畔 乔林 徐立波 于元旗 韩永辉 HU Pan;QIAO Lin;XU Libo;YU Yuanqi;HAN Yonghui(Information and Communication Branch of State Grid Liaoning Electric Power Supply Co.,Ltd.,Shenyang 110006,China)

机构地区：[1]国网辽宁省电力有限公司信息通信分公司

出　　处：《计算机工程》2019年第11期269-274,共6页Computer Engineering

基　　金：国家重点研发计划(2017YFC0804714)

摘　　要：目前多数跟踪算法采用尺度遍历穷搜索策略应对目标的尺度变化,其跟踪性能和效率不佳。针对此问题,基于特定目标提议框提出一种自适应跟踪算法。对目标提议框生成算法进行改进,融入跟踪目标的尺度和位置信息,得到特定目标提议框并获取其特征。为确保跟踪的连续性,将自适应支持向量机作为跟踪模型,对特定目标提议框进行评分,得到目标位置。对均匀采样样本和特定目标提议框正负样本分类,进行模型更新。在OTB100数据库上进行对比实验,结果表明,与CNN-SVM、DeepSRDCF等算法相比,该算法能较好地适应目标的尺度变化和形变,有效提高跟踪效率。Currently,most tracking algorithms use scale-traversing poor search strategy to cope with the scale changes of targets,but the tracking performance and efficiency are not satisfying.Therefore,we propose an adaptive tracking algorithm base on specific object proposal.First,we adopt the scale and position information of the tracking target into the object proposal generation algorithm,so as to obtain the specific object proposal and its features.Then,to ensure the tracking continuity,we use the adaptive Support Vector Machine(SVM)as tracking model to score the specific object proposal,thus obtaining the target.Finally,we update the model by classifying the positive and negative samples of specific object proposal and the uniformly-sampled samples.Comparative experiments on OTB100 dataset show that compared with CNN-SVM and DeepSRDCF algorithms,the proposed algorithm can better adapt to the scale changes and deformation of the target,and improve tracking efficiency.

关 键 词：特定目标提议框 自适应跟踪 视觉目标跟踪 深度学习 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]