摘要
核相关滤波KCF(kernel correlation filter)算法存在的尺度不能自适应及目标丢失后不能重新自动框选的问题,提出一种基于KCF框架的改进算法;该算法基于KCF框架,引入双重尺度估计策略,实现目标尺度自适应的同时又提高跟踪速度;同时还引入峰值旁瓣比PSR(peak to side lobe ratio)来进行遮挡判断,引入加速稳健特征SURF(speeded up robust features)特征来对目标进行自动重新匹配并框选。实验结果表明,提出的改进算法相比于原算法能够实现目标跟踪尺度自适应,并且在目标丢失后自动重新匹配框选目标并继续跟踪,实现对目标的长时间稳定跟踪,使其在铅鱼跟踪上有较好的效果。
Aiming at the problem that KCF algorithm can‘t be scale adaptive and the frame selection can’t be automatically restored after the target is lost,an improved algorithm based on KCF framework is proposed;Based on the KCF framework,this algorithm introduces a Dual-scale Estimation Strategy.It can realize the target scale adaptation and improve the tracking speed at the same time;Meanwhile,peak to sidelobe ratio(PSR)is introduced to judge whether there is occlusion in the target,and introduce SURF(speeded up robust features)feature to automatically re-match and frame select targets.The experimental results show that the improved algorithm can achieve object tracking scale adaptation compared with the original algorithm,automatically rematch frame select and tracking of targets,after the target is lost and continue tracking,achieve long-term stable tracking of the target,which has a better effect on Elliptic-type weight tracking.
作者
王挥华
王剑平
张果
欧阳鑫
Wang Huihua;Wang Jianping;Zhang Guo;Ouyang Xin(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650000,China)
出处
《电子测量技术》
2020年第8期117-124,共8页
Electronic Measurement Technology
基金
国家重点研发计划(2017YFB0306405)
云南省重点研发项目(2018BA070)
国家自然基金(61364008)
昆明理工大学复杂工业控制学科方向团队建设计划项目资助
作者简介
王挥华,硕士研究生,要研究方向为计算机视觉,复杂工业过程控制。E-mail:763488067@qq.com;通信作者:王剑平,博士,副教授,主要研究方向为复杂工业过程控制。E-mail:kmustwjp@126.com;张果,博士,副教授,主要研究方向为智能测控系统、工业通信网络。E-mail:21426717@qq.com;欧阳鑫,博士,主要研究方向为信号处理,通信工程。E-mail:812120164@qq.com
