DCFNet算法的物体长时跟踪解决方案
|
摘要
提出了一种基于DCFNet算法的长时跟踪解决方案。首先添加了模版更新策略,可以避免在物体发生遮挡时不必要的模版更新;其次添加了全局搜索策略,当跟踪算法在局部丢失目标物后,全局搜索策略会起作用,进行快速全局搜索,找到目标重新开始跟踪。全局搜索策略采用级联分类器的思想,保证了该策略的执行速度。虽然与DCFNet相比,本文提出的方法在平均速度上略有下降,但是在性能上有所提升,并且在DCFNet跟踪失败时,能够再次检测到目标物的位置持续跟踪下去。
In the paper,a solution for long-term tracking is proposed based on DCFNet.First,a template update strategy is added to avoid unnecessary template updates when objects are occluded.At the same time,aglobal search strategy is added.When the tracking algorithm loses the target locally,the global search strategy will work,and a fast global search is performed to find the target to start tracking again.The global search strategy uses the idea of a cascaded classifier to ensure the execution speed of the strategy.Compared with DCFNet,the proposed method has a slight decrease in average speed,but it has increased in performance,and when the DCFNet tracking fails,the position of the target can be detected again and tracked continuously.
In the paper,a solution for long-term tracking is proposed based on DCFNet.First,a template update strategy is added to avoid unnecessary template updates when objects are occluded.At the same time,aglobal search strategy is added.When the tracking algorithm loses the target locally,the global search strategy will work,and a fast global search is performed to find the target to start tracking again.The global search strategy uses the idea of a cascaded classifier to ensure the execution speed of the strategy.Compared with DCFNet,the proposed method has a slight decrease in average speed,but it has increased in performance,and when the DCFNet tracking fails,the position of the target can be detected again and tracked continuously.
引文
[1]Wang N,Shi J,Yeung D Y,et al.Understanding and diagnosing visual tracking systems[C]//ICCV,2015.
[2]Henriques J F,Rui C,Martins P,et al.High-speed Tracking with Kernelized Correlation Filters[C]//IEEETPAMI,2015.
[3]Wang Q,Gao J,Xing J,et al.DCFNet:Discriminant Correlation Filters Network for Visual Tracking[C]//ICIP,2017.
[4]Zhu G,Porikli F,Li H.Beyond local search:Tracking objects everywhere with instance-specific proposals[C]//CVPR,2016.
[5]Kalal Z,Mikolajczyk K,Matas J.Tracking-learning-detection[C]//IEEE TPAMI,2012.
[2]Henriques J F,Rui C,Martins P,et al.High-speed Tracking with Kernelized Correlation Filters[C]//IEEETPAMI,2015.
[3]Wang Q,Gao J,Xing J,et al.DCFNet:Discriminant Correlation Filters Network for Visual Tracking[C]//ICIP,2017.
[4]Zhu G,Porikli F,Li H.Beyond local search:Tracking objects everywhere with instance-specific proposals[C]//CVPR,2016.
[5]Kalal Z,Mikolajczyk K,Matas J.Tracking-learning-detection[C]//IEEE TPAMI,2012.