摘要
传统目标跟踪算法很难在户外视频中进行行人实时跟踪及人流量统计,针对该问题,提出一种联合邻帧匹配与卡尔曼滤波器的多目标跟踪算法。该算法首先以方向梯度直方图(HOG)为特征的自适应推进(Adaboost)分类器进行行人人头检测,进而采用卡尔曼滤波进行行人轨迹预测,并应用邻帧匹配算法对该目标运动轨迹进行校正。在邻帧匹配法中,新颖地将跟踪过程中的目标匹配问题转化为基于匈牙利算法的指派问题。因邻帧匹配法能更好地区分不同待匹配目标的特征,较传统跟踪算法,能更好地避免误匹配、漏匹配等问题,提高了目标跟踪的准确率,实验证明,该监控系统准确率高,且能满足扶梯监控系统应用场景下的鲁棒性、实时性要求。
In order to segment pedestrians in the crowded scene of real time video,a pedestrian tracking algorithm is proposed,which combine neighborhood frame matching and kalman filtering. Firstly,gradient histogram( HOG) combined with adaptive boosting( Adaboost) were used to detect the passenger heads in the scene. Then,neighborhood frame matching method combined with Kalman filter were used for multi-target tracking. What's more,the target matching problem is transformed into assignment problem based on Hungary algorithm. For the adjacent frames matching method can better distinguish the characteristics of different target,compared to traditional tracking algorithms,it can better avoid false matching and miss matching,improves the accuracy of target tracking.Experiments show that the algorithm is of high accuracy,and can meet the requirements of robustness and real-time performance in the application of video monitoring system.
引文
1 Fukunaga K,Hostetler L.The estimation of the gradient of a density function,with applications in pattern recognition[J].IEEE Transactions on Information Theory,1975,21(1):32-40
2 Yang C,Duraiswami R,Davis L.Efficient mean-shift tracking via a new similarity measure[C]//IEEE Computer Society Conference on Computer Vision&Pattern Recognition.San Diego:IEEE,2005:176-183
3 Nouar O D,Ali G,Rapha3l C.Improved object tracking with camshift algorithm[C]//IEEE International Conference on Acoustics Speech&Signal Processing.Toulouse:IEEE,2006:439-446
4 Mukherjee S,Saha B,Jamal I,et al.A novel framework for automatic passenger counting[C]//IEEE International Conference on Image Processing.Brussels:IEEE,2011:2969-2972
5 Isard M,Blake A.Condensation-conditional density propagation for visual tracking[J].International Journal of Computer Vision,1998,29(1):5-28
6 Kass M,Witkin A,Terzopoulos D.Snakes active contour models[J].International Journal of Computer Vision,1988,1(4):321-331
7刘婕.复杂场景多特征融合粒子滤波目标跟踪[D].重庆:重庆理工大学,2015Liu Jie.Multi-feature fusion target tracking based on particle filter in complex scene[D].Chongqing:Chongqing University of Technology,2015
8 Henriques J F,Rui C,Martins P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015,37(3):583-596
9 Stepanov O A.Kalman filtering:Past and present.An outlook from Russia[J].Gyroscopy and Navigation,2011,2(2):99-110
10 Salmond D.Target tracking:Introduction and Kalman tracking filters[C]//IEEE Target Tracking:Algorithms&Applications,Enschede.Netherlands:IET,2002:537-543
11王书朋.视频目标跟踪算法研究[D].西安:西安电子科技大学,2014Wang Shupeng.Research on methods of visual object tracking[D].Xi'an:Xidian University,2014
12张英.基于视频的实时人数统计关键技术的研究[D].杭州:杭州电子科技大学,2014Zhang Ying.Research on key techniques of real-time people counting based on video[D].Hangzhou:Hangzhou Electronic Science and Technology University,2014
13杜启亮,黎浩正,田联房.基于Adaboost和本模型的手扶电梯出入口视频监控方法[J].计算机应用,2017,37(9):2610-2616Du Qiliang,Li Haozheng,Tian Lianfang.Video monitoring method of escalator entrance area based on Adaboost and codebook model[J].Journal of Computer Applications,2017,37(9):2610-2616
14张旭辉,朱宏辉,郑启忠.最优指派问题匈牙利算法的探讨与C++实现[J].物流技术,2004(5):67-69Zhang Xuhui,Zhu Honghui,Zheng Qizhong.Hungary algorithm concerning best allotting and C++[J].Logistics Technology,2004(5):67-69