基于动态结构数组的多目标跟踪初始化方法
|
摘要
提出一种通过构建动态结构数组来自动初始化多目标跟踪(multi-target tracking, MTT)的方法,将检测算法提取的目标区域信息打包成数据集进行独立存储;构造区域信息的结构数组,生成可供跟踪算法调用的mat文件;运行目标跟踪算法,自动建立初始多目标轨迹;利用连续帧判别并综合历史检测信息对新出现的目标动态初始化.实验结果表明:该方法在MTT全局初始化和新出现目标动态初始化方面均具有良好的鲁棒性,并且能实现任意2个独立的多目标检测和MTT算法的自动衔接.
This paper proposes a method for automatically initializing MTT(multi-target tracking) by constructing dynamic structure arrays. The target region information extracted by the detection algorithm is packaged into a data set and independently stored. A structure array containing the region information is constructed to generate a mat file which can be called by the tracking algorithm. The initial tracktories of multi-target can be automatically established by runnig the tracker. The sequential frames discrimination and integrated information of the historical detections are utilized to dynamically initialize the newly appearing targets. The experimental results prove the great robustness of the proposed method in terms of the global initialization of MTT and the dynamic initialization of new targets in the tracking process. Furthermore, the method can automaticly connect any two independent multi-target detections and MTT algorithms.
This paper proposes a method for automatically initializing MTT(multi-target tracking) by constructing dynamic structure arrays. The target region information extracted by the detection algorithm is packaged into a data set and independently stored. A structure array containing the region information is constructed to generate a mat file which can be called by the tracking algorithm. The initial tracktories of multi-target can be automatically established by runnig the tracker. The sequential frames discrimination and integrated information of the historical detections are utilized to dynamically initialize the newly appearing targets. The experimental results prove the great robustness of the proposed method in terms of the global initialization of MTT and the dynamic initialization of new targets in the tracking process. Furthermore, the method can automaticly connect any two independent multi-target detections and MTT algorithms.
引文
[1] WANG Qing, CHEN Feng, XU Wenli, et al. Object tracking with joint optimization of representation and classification [J]. IEEE Trans Circuits Syst Video Technol, 2015, 25(4): 638-650.
[2] 王轶, 方英武. 基于DST-PCR5多目标自适应视觉跟踪方法 [J]. 计算机应用研究, 2018, 35(2): 632-636.
[3] YUAN Yuan, LU Yuwei, WANG Qi. Tracking as a whole: multi-target tracking by modeling group behavior with sequential detection [J]. IEEE Trans Intell Transp Syst, 2017, 18(12): 3339-3349.
[4] BAE S H, YOON K J. Confidence-based data association and discriminative deep appearance learning for robust online multi-object tracking [J]. IEEE Trans Pattern Anal Mach Intell, 2018, 40(3): 595-610.
[5] XIANG Jun, SANG Nong, HOU Jianhua, et al. Hough forest-based association framework with occlusion handling for multi-target tracking [J]. IEEE Signal Process Lett, 2016, 23(2): 257-261.
[6] MILAN A, SCHINDLER K, ROTH S. Multi-target tracking by discrete-continuous energy minimization [J]. IEEE Trans Pattern Anal Mach Intell, 2016, 38(10): 2054-2068.
[7] IDREES H, SOOMRO K, SHAH M. Detecting humans in dense crowds using locally-consistent scale prior and global occlusion reasoning [J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(10): 1986-1988.
[8] BAE S H, YOON K J. Robust online multiobject tracking with data association and track management [J]. IEEE Trans Image Process, 2014, 23(7): 2820-2833.
[9] HENRIQUES J S, CASEIRO R, MARTINS P, et al. High-speed tracking with kernelized correlation filters [J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(3): 583-596.
[10] XU Yulong, WANG Jiabao, LI Yang, et al. One-step backtracking for occlusion detection in real-time visual tracking [J]. Electron Lett, 2017, 53(5): 318-320.
[2] 王轶, 方英武. 基于DST-PCR5多目标自适应视觉跟踪方法 [J]. 计算机应用研究, 2018, 35(2): 632-636.
[3] YUAN Yuan, LU Yuwei, WANG Qi. Tracking as a whole: multi-target tracking by modeling group behavior with sequential detection [J]. IEEE Trans Intell Transp Syst, 2017, 18(12): 3339-3349.
[4] BAE S H, YOON K J. Confidence-based data association and discriminative deep appearance learning for robust online multi-object tracking [J]. IEEE Trans Pattern Anal Mach Intell, 2018, 40(3): 595-610.
[5] XIANG Jun, SANG Nong, HOU Jianhua, et al. Hough forest-based association framework with occlusion handling for multi-target tracking [J]. IEEE Signal Process Lett, 2016, 23(2): 257-261.
[6] MILAN A, SCHINDLER K, ROTH S. Multi-target tracking by discrete-continuous energy minimization [J]. IEEE Trans Pattern Anal Mach Intell, 2016, 38(10): 2054-2068.
[7] IDREES H, SOOMRO K, SHAH M. Detecting humans in dense crowds using locally-consistent scale prior and global occlusion reasoning [J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(10): 1986-1988.
[8] BAE S H, YOON K J. Robust online multiobject tracking with data association and track management [J]. IEEE Trans Image Process, 2014, 23(7): 2820-2833.
[9] HENRIQUES J S, CASEIRO R, MARTINS P, et al. High-speed tracking with kernelized correlation filters [J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(3): 583-596.
[10] XU Yulong, WANG Jiabao, LI Yang, et al. One-step backtracking for occlusion detection in real-time visual tracking [J]. Electron Lett, 2017, 53(5): 318-320.