基于核相关滤波的TLD跟踪算法研究
|
摘要
结合跟踪学习检测(TLD)与核相关滤波算法,开发了一种基于核相关滤波的TLD跟踪算法。首先对核相关滤波进行多尺度改进,以适应目标的尺度变化,其次在TLD的跟踪模块中引入改进后的核相关滤波算法,以提高运行效率和跟踪的鲁棒性,最后在TLD的学习模块中引入自适应学习参数,及时更新核相关滤波器,以提高系统的跟踪精度。实验结果表明,与核相关滤波和TLD算法相比,本文所研究的算法在复杂背景下的目标跟踪具有较好的鲁棒性与适应性,提高了算法的运行效率和跟踪的稳定性。
Combining the tracking learning detection(TLD) algorithm with kernel correlation filtering algorithm,this paper develops a TLD tracking algorithm based on kernel correlation filtering,firstly performs multi-scale improvement on the kernel correlation filtering to adapt to the target scale change,secondly introduces an improved kernel correlation filtering algorithm in the TLD tracking module to improve the operational efficiency and the robustness of the tracking,finally leads in the adaptive learning parameters to the TLD learning module,updates the kernel correlation filter to improve the tracking accuracy of the system.The experimental results show that compared with the kernel correlation filtering and TLD algorithms,the algorithm studied in this paper has better robustness and adaptability to target tracking in complex background,and improves the operating efficiency and tracking stability of the algorithm.
Combining the tracking learning detection(TLD) algorithm with kernel correlation filtering algorithm,this paper develops a TLD tracking algorithm based on kernel correlation filtering,firstly performs multi-scale improvement on the kernel correlation filtering to adapt to the target scale change,secondly introduces an improved kernel correlation filtering algorithm in the TLD tracking module to improve the operational efficiency and the robustness of the tracking,finally leads in the adaptive learning parameters to the TLD learning module,updates the kernel correlation filter to improve the tracking accuracy of the system.The experimental results show that compared with the kernel correlation filtering and TLD algorithms,the algorithm studied in this paper has better robustness and adaptability to target tracking in complex background,and improves the operating efficiency and tracking stability of the algorithm.
引文
[1] LIANG N,WU G,KANG W,et al.Real-time long-term tracking with prediction-detection-correction[J].IEEE Transactions on Multimedia,2018,20(9):2289-2302.
[2] 梁宵.基于检测区域自动调整的TLD目标跟踪研究[D].长春:吉林大学,2016.
[3] KALAL Z,MIKOLAJCZYK K,MATAS J.Tracking-learning-detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2012,34(7):1409-1422.
[4] 刘万军.遮挡判别下的多尺度相关滤波跟踪算法[J].中国图象图形学报,2018,23(12):1789-1800.
[5] 孟煜.基于TLD的多目标快速跟踪算法研究[D].阜新:辽宁工程技术大学,2015.
[6] HENRIQUES J F,RUI C,MARTINS P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.
[2] 梁宵.基于检测区域自动调整的TLD目标跟踪研究[D].长春:吉林大学,2016.
[3] KALAL Z,MIKOLAJCZYK K,MATAS J.Tracking-learning-detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2012,34(7):1409-1422.
[4] 刘万军.遮挡判别下的多尺度相关滤波跟踪算法[J].中国图象图形学报,2018,23(12):1789-1800.
[5] 孟煜.基于TLD的多目标快速跟踪算法研究[D].阜新:辽宁工程技术大学,2015.
[6] HENRIQUES J F,RUI C,MARTINS P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.