摘要
核相关滤波(KCF)已广泛应用于目标跟踪,但如何调整KCF的正则项来提高跟踪精度一直是一个公开问题。本算法针对目标遮挡、尺度变化等复杂场景,提出了一种基于重检测和目标遮挡判定的相关滤波跟踪算法。首先在KCF的框架下融入了遮挡判定模块,实施遮挡判决;若无遮挡则正常跟踪,否则启动重检测模块。其次,通过在可信度大的几个位置设置锚点,设计了一种重检测模块,提高了目标的跟踪准确率,也避免了做全局搜索带来的计算资源浪费。最后,通过采用尺度估计模块来预测目标框的真实大小,减小了因目标形变带来的模型损耗。通过仿真测试和应用实例表明,当跟踪目标发生遮挡和尺度变化时,所提算法的精确度和鲁棒性均有一定的提升。
Kernel correlation filter(KCF) has been widely used for target tracking, but how to adjust the KCF regular term to improve tracking accuracy has always been a public issue. Aiming at the complex tracking environment such as target occlusion and scale change, we propose a correlation filter tracking algorithm based on re-detection and target occlusion decision. Firstly, the algorithm adds the occlusion decision module under the framework of KCF. If there is no occlusion, it will track normally. Otherwise, the re-detection module will be started. The detection points will be anchored at high reliability locations. The module improves the tracking accuracy and avoids the waste of computing resources caused by global search. Finally, the scale estimation module is applied to predict the true size of the target frame and reduce the model loss caused by the deformation of the object. Simulation tests and application examples show that the accuracy and robustness of the proposed algorithm are improved to some extent when the tracking target is occluded or scale change.
作者
易宇凡
瞿少成
陶林
Yi Yufan;Qu Shaocheng;Tao Lin(College of Physical Science and Technology,Central China Normal University,Wuhan 430079,China)
出处
《电子测量技术》
2020年第7期93-96,共4页
Electronic Measurement Technology
基金
国家自然科学基金资助项目(61673190/F030101)
教育部中央高校探索创新项目(CCNU18TS042)
武汉市洪山区产学研项目资助
关键词
核相关滤波(KCF)
单目标跟踪
目标遮挡辨别
重检测策略
kernel correlation filter(KCF)
single target tracking
target occlusion discrimination
re-detection strategy
