一种基于SVM的核相关跟踪算法  

An algorithm of kernelized correlation tracking with SVM

作　　者：袁康 魏大鹏[2] 赵从梅 傅顺 YUAN Kang;WEI Da-peng;ZHAO Cong-mei;FU Shun(College of Computer Science and Technology, Chongqing University of Posts ＆ Telecommunications, Chongqing 400065, China;Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences, Chongqing 400714, China)

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065 [2]中国科学院重庆绿色智能技术研究院,重庆400714

出　　处：《传感器与微系统》2018年第5期138-143,共6页Transducer and Microsystem Technologies

摘　　要：基于相关滤波器的跟踪方法在准确度和鲁棒性上取得了突出优势,但仍需要提高整体的跟踪性能。针对传统单目标的核相关滤波器跟踪算法在目标尺度变化和产生遮挡的跟踪中存在的问题,提出了一种结合支持向量机(SVM)检测器的多尺度相关滤波器算法。通过在核矩阵中引入尺度因子来提高相关滤波器处理尺度变换的性能,训练了一个在线SVM检测器,当目标发生遮挡时,能够重新获取目标,同时自适应调整模型学习率。通过与其他5种优秀跟踪算法进行实验比较,结果表明:方法能够广泛应用于目标跟踪领域,对目标进行准确地估计并有效处理目标的遮挡问题。Although correlation filter-based tracking method achieve competitive results both on accuracy and robustness ,there is still a need to improve the overall tracking capability. Aiming at problem of kernelized correlation fiher-hased tracking algorithm in scale target variation and occlusion, present a muhi-scale correlation tracker algorithru combined with support vector machine （ SVM ） detector to solve the above problems. By introducing the scale factor into the kernel matrix to improve the performance of correlation filter processing scale transform. Train an online SVM detector to retrieve the target when the target occluded, and adaptively adjust the learning rate of the model. By comparing with the other five outstanding tracking algorithms. Experimental results show that the proposed approach can estimate the object state accurately and handle the object occlusion problem effectively.

关 键 词：目标跟踪 支持向量机 多尺度 相关滤波器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]