摘要
针对低照度、背景杂乱等较差条件下的目标跟踪不稳定问题,论文研究了如何自适应利用多模态信息有效并快速地实现目标的持续稳健跟踪.为此,提出一种基于可靠相关度的实时多模态目标跟踪方法.给定空域上、时序上均对齐的热红外和可见光视频序列,使用在线训练的相关性滤波器对视频帧进行滤波,得到各个模态下的目标置信度图.然后,把目标置信度图的峰值主副比作为模态可靠性的度量,以此选择最为可靠的目标置信度图,得到最优的跟踪结果.同时,提出一种简单有效的多模态更新方法,使得不同模态的目标模型能够适应目标外观的变化,且避免噪声的影响.为了全面地评价论文方法,构建了一个包含6组多模态视频,其中包含了多种挑战因素,如低照度、热交叉等.实验结果表明,论文方法比其他方法提高了25%左右的跟踪精度.
The paper investigated how to adaptively employ multimodal information to achieve robust object tracking in some challenging scenarios,such as low illumination,background clutter and so on,which easily led to model drift in visual tracking.Therefore,we proposed a real-time multimodal tracking method based on selecting reliable correlation filter.Given the aligned thermal and gray-scale video pairs,the online trained correlation filter was firstly employed in each modality to obtain the confidence map.Secondly,the peak-to-sidelobe ratio(PSR)of the confidence map was treaded as the model reliability measurement,which the optimal tracking result could be obtained by selecting most reliable modality.Moreover,to adapt to the variation of object appearance and alleviate the effects of noises,a simple yet effective update scheme was proposed.Finally,to evaluate the proposed method comprehensively,we built a multimodal video data set which includes several challenging factors,such as low illumination and thermal crossover.The experimental results suggested that the proposed method outperforms the other tracking methods about 25%in accuracy.
引文
[1]HENRIQUES J F,CASEIRO R,MARTINS P,et al.High-speed tracking with kernelized correlation filters[J].Pattern Analysis and Machine Intelligence,IEEE Transactions on,2015,37(3):583-596.
[2]ZHANG K,ZHANG L,LIU Q,et al.Fast visual tracking via dense spatio-temporal context learning[C]//Computer Vision-ECCV,2014:127-141.
[3]LING H,BAI L,BLASCH E,et al.Robust infrared vehicle tracking across target pose change using L1regularization[C]//Information Fusion(FUSION),13th Conference on,IEEE,2010:1-8.
[4]BAI Y,TANG M.Robust tracking via weakly supervised ranking svm[C]//Computer Vision and Pattern Recognition(CVPR),IEEE Conference on,IEEE,2012:1854-1861.
[5]MEI X,LING H.Robust visual tracking using l1 minimization[C]//Computer Vision,IEEE 12th International Conference on,IEEE,2009:1436-1443.
[6]ZHANG T,GHANEM B,LIU S,et al.Robust visual tracking via multi-task sparse learning[C]//Computer Vision and Pattern Recognition(CVPR),IEEE Conference on,IEEE,2012:2042-2049.
[7]BAO C,WU Y,LING H,et al.Real time robust l1tracker using accelerated proximal gradient approach[C]//Computer Vision and Pattern Recognition(CVPR),IEEE Conference on,IEEE,2012:1830-1837.
[8]CHEN S,ZHU W,LEUNG H.Thermo-visual video fusion using probabilistic graphical model for human tracking[C]//Circuits and Systems,ISCAS,IEEE International Symposium on,IEEE,2008:1926-1929.
[9]程咏梅,王进行,魏坤,等.基于加权mean-shift可见光/热红外双通道目标跟踪[J].光子学报,2010,39(2):357-363.
[10]WU Y,BLASCH E,CHEN G,et al.Multiple source data fusion via sparse representation for robust visual tracking[C]//Information Fusion,Proceedings of the 14th International Conference on,IEEE,2011:1-8.
[11]LIU H P,SUN F C.Fusion tracking in color and infrared images using joint sparse representation[J].Science China Information Sciences,2012,55(3):590-599.
[12]CONAIRE C O,O’CONNOR N E,SMEATON A.Thermo-visual feature fusion for object tracking using multiple spatiogram trackers[J].Machine Vision and Applications,2008,19(5/6):483-494.
[13]LEYKIN A,HAMMOUD R.Pedestrian tracking by fusion of thermal-visible surveillance videos[J].Machine Vision and Applications,2010,21(4):587-595.