适合长时跟踪的自适应相关滤波算法  

Adaptive Correlation Filtering Algorithm Suitable for Long-Term Tracking

作　　者：肖逸清 葛洪伟 Xiao Yiqing;Ge Hongwei(Jiangsu Provincial Engineering Laboratory of Pattern Recognition and Computational Intelligence(Jiangnan University),Wuxi 214122;School of Internet of Things,Jiangnan University,Wuxi 214122)

机构地区：[1]江苏省模式识别与计算智能工程实验室(江南大学),无锡214122 [2]江南大学物联网工程学院,无锡214122

出　　处：《计算机辅助设计与图形学学报》2020年第1期121-129,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：江苏省研究生创新计划项目(KYLX16_0781);江苏高校优势学科建设工程资助项目

摘　　要：针对在长时跟踪中,快速运动、遮挡等复杂情况很容易引起模板漂移,导致跟踪失败的问题,提出一种适合长时跟踪的自适应相关滤波算法.首先融合HOG特征、CN特征和灰度特征,在增强特征判别力的同时,结合EdgeBoxes生成检测建议并找到最优建议,实现跟踪器尺度与纵横比的自适应;然后利用高置信度跟踪结果来避免模板被破坏,将目标移动速度与边缘组数结合起来形成一种新的自适应更新率,并对每一帧目标框的尺度进行校正;最后在跟踪失败的情况下,应用增量学习检测器以滑动窗口的方式恢复目标位置.在标准测试集上与基于相关滤波的7种算法进行对比,实验表明,该算法在精确度和成功率上均取得较优效果.In view of the problem that fast motion,occlusion and other complex conditions can easily cause template drift and the track failure in long-term tracking,an adaptive correlation filtering algorithm suitable for the long time tracking is proposed.First of all,this paper integrates HOG feature,CN feature and gray feature to enhance the discriminant power of features.While it combines EdgeBoxes to generate detection proposals to find the optimal proposal to realize the adaptive scale and aspect ratio,this paper uses high confidence tracking results to avoid the template being destroyed.The target’s speed and the number of edge groups are combined to form a new adaptive update rate,and this paper corrects the scale of the target box for each frame.Finally,in the case of trace failure,the incremental learning detector is applied to restore the target position by sliding window.Compared with other 7 algorithms based on correlation filtering on the standard test set,the experimental results show that the proposed algorithm achieves better results in accuracy and success rate.

关 键 词：核相关滤波 长时跟踪 模型更新 目标重检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]