Fast Compressive Tracking Algorithm Combining Feature Selection with Secondary Localization
-
摘要: 针对压缩跟踪算法易受遮挡影响和模型更新比较盲目的问题，提出结合特征筛选与二次定位的快速压缩跟踪算法（Fast compressive tracking algorithm combining feature selection with secondary localization，FSSL-CT）.首先，对全局区域划分子区域，从中提取压缩特征，根据正、负样本估计出各个压缩特征在正、负类中的分布; 然后，使用自适应学习率结合正类更新阈值对分类器模型进行更新; 最后，将跟踪分为两个阶段，每个阶段在对应的搜索区域内采集候选样本，并从全部特征中筛选出部分优质特征加权构建分类器，通过分类候选样本最终完成目标跟踪.在8个公共测试序列和4个自制序列中与最近提出的两个代表性算法进行比较，本文算法在大多数测试序列中都具有最高的跟踪成功率和最低的平均中心误差，平均处理速度可以达到3.04毫秒/帧.实验结果表明，本文算法具有更好的抵抗短时遮挡的能力，更高的准确性和鲁棒性，以及良好的实时性.Abstract: As the traditional compressive tracking algorithm fails to track targets stably under occlusive condition and update model accurately, a fast tracking algorithm combining feature selection with secondary localization based on compressive tracking (FSSL-CT) is proposed. Firstly, compressive features are extracted from sub-regions partitioned from the global region, and the distributions of each compressive feature in positive and negative classes are estimated. Secondly, the classifier model is updated utilizing the method of adaptive learning rate and positive class update threshold. Finally, the tracking stage is divided into two procedures. In each procedure, some candidate samples are collected in the given searching region, and partial high quality features are selected from all the features and weighted to construct a classifier, then, the candidate samples are classified by the classifier. After that, the target tracking is achieved. Compared with two state-of-the-art algorithms on 8 public testing sequences and 4 private sequences, the FSSL-CT algorithm is proved to have the highest tracking success ratio and the lowest average central error in most of the sequences, and the average processing speed could achieve 3.04 milliseconds per frame. It is tested that the proposed FSSL-CT algorithm has a better capacity of resisting short-time occlusion and running in real-time, higher accuracy and robustness than the two state-of-the-art algorithms.
-
表 1 各算法跟踪成功率比较
Table 1 Comparison of the tracking success rate of different algorithms
视频序列 FSSL-CT (\%) CT/原文(\%) FCT/原文(\%) Jogging 95.76 22.47/— 22.14/— Girl 85.00 65.40/78 52.20/— Shaking 81.91 85.20/92 93.42/97 FleetFace 90.24 76.80/— 67.60/— KiteSurf 98.80 70.23/68 57.14/— Tiger1 71.75 64.40/78 52.82/52 Sylvester 86.39 80.29/75 69.29/77 FaceOcc2 90.02 96.55/100 89.16/— Hat 98.00 83.04/— 89.90/— DayLight1 100.00 48.90/— 76.55/— DayLight2 100.00 91.78/— 100.00/— Night 100.00 77.56/— 100.00/— 
下载: 导出CSV
表 2 各算法平均中心误差比较(像素)
Table 2 Comparison of mean center error of different algorithms (pixel)
视频序列 FSSL-CT CT/原文 FCT/原文 Jogging 12.07 94.48/— 93.59/— Girl 5.66 8.31/21 12.74/— Shaking 11.78 13.20/9 10.70/14 FleetFace 16.70 52.50/— 54.72/— KiteSurf 2.24 6.12/9 13.93/— Tiger1 22.40 25.35/10 35.59/23 Sylvester 9.94 16.56/9 20.70/9 FaceOcc2 12.57 11.78/10 16.10/— Hat 8.81 18.09/— 11.19/— DayLight1 9.89 25.70/— 19.54/— DayLight2 6.69 15.91/— 8.90/— Night 6.16 14.00/— 9.53/—
下载: 导出CSV
表 3 各算法计算次数及相关参数(次)
Table 3 Calculation times and related parameters of different algorithms (times)
参数 FSSL-CT CT/原文 FCT/原文 m1 81 1961 121 n1 40 50 100 m2 81 0 317 n2 80 0 100 N 9720 98050 43800
下载: 导出CSV
表 4 各算法运行速度比较(毫秒/帧)
Table 4 Comparison of operation speed of different algorithms (ms/frame)
视频序列 FSSL-CT CT FCT Jogging 3.04 18.10 11.34 Girl 2.63 18.85 11.20 Shaking 3.31 19.95 12.39 FleetFace 3.09 21.35 12.45 KiteSurf 2.97 21.39 11.89 Tiger1 3.41 21.57 12.85 Sylvester 2.72 20.86 12.36 FaceOcc2 3.06 21.38 12.75 Hat 2.88 19.24 11.43 DayLight1 3.17 20.26 11.49 DayLight2 3.26 21.43 11.92 Night 2.98 21.36 11.76 平均处理时间 3.04 20.48 11.99
下载: 导出CSV
-
[1] Comaniciu D, Ramesh V, Meer P. Kernel-based object tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25(5): 564-577 doi: 10.1109/TPAMI.2003.1195991 [2] Bradski G R. Real time face and object tracking as a component of a perceptual user interface. In: Proceedings of the 4th IEEE Workshop on Applications of Computer Vision. Princeton, USA: IEEE, 1998. 214-219 [3] Huang S L, Hong J X. Moving object tracking system based on camshift and Kalman filter. In: Proceedings of the 2011 International Conference on Consumer Electronics, Communications and Networks. Xianning, China: IEEE, 2011. 1423-1426 [4] Vadakkepat P, Liu J. Improved particle filter in sensor fusion for tracking randomly moving object. IEEE Transactions on Instrumentation and Measurement, 2006, 55(5): 1823-1832 doi: 10.1109/TIM.2006.881569 [5] 王智灵, 陈宗海, 徐萧萧, 吴亮.基于蛙眼视觉特性的运动目标模糊化区域理解跟踪方法.自动化学报, 2009, 35(8): 1048-1054 doi: 10.3724/SP.J.1004.2009.01048Wang Zhi-Ling, Chen Zong-Hai, Xu Xiao-Xiao, Wu Liang. A fuzzy region understanding tactic for object tracking based on frog's vision characteristic. Acta Automatica Sinica, 2009, 35(8): 1048-1054 doi: 10.3724/SP.J.1004.2009.01048 [6] 李智勇, 何霜, 刘俊敏, 李仁发.基于蛙眼R3细胞感受野模型的运动滤波方法.自动化学报, 2015, 41(5): 981-990 http://www.aas.net.cn/CN/abstract/abstract18672.shtmlLi Zhi-Yong, He Shuang, Liu Jun-Min, Li Ren-Fa. Motion filtering by modelling R3 cell's receptive field in frog eyes. Acta Automatica Sinica, 2015, 41(5): 981-990 http://www.aas.net.cn/CN/abstract/abstract18672.shtml [7] 刘伟峰, 柴中, 文成林.基于随机采样的多量测目标跟踪算法.自动化学报, 2013, 39(2): 168-178 http://www.aas.net.cn/CN/abstract/abstract17858.shtmlLiu Wei-Feng, Chai Zhong, Wen Cheng-Lin. Multi-measurement target tracking by using random sampling approach. Acta Automatica Sinica, 2013, 39(2): 168-178 http://www.aas.net.cn/CN/abstract/abstract17858.shtml [8] Avidan S. Support vector tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2004, 26(8): 1064-1072 doi: 10.1109/TPAMI.2004.53 [9] Yang F, Lu H C, Yang M H. Robust superpixel tracking. IEEE Transactions on Image Processing, 2014, 23(4): 1639-1651 doi: 10.1109/TIP.2014.2300823 [10] Collins R T, Liu Y X, Leordeanu M. Online selection of discriminative tracking features. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10): 1631-1643 doi: 10.1109/TPAMI.2005.205 [11] Babenko B, Yang M H, Belongie S. Robust object tracking with online multiple instance learning. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(8): 1619-1632 doi: 10.1109/TPAMI.2010.226 [12] Zhang T Z, Ghanem B, Liu S, Ahuja N. Robust visual tracking via multi-task sparse learning. In: Proceedings of the 2012 IEEE Conference on Computer Vision and Pattern Recognition. Providence, USA: IEEE, 2012. 2042-2049 [13] Zhong W, Lu H C, Yang M H. Robust object tracking via sparse collaborative appearance model. IEEE Transactions on Image Processing, 2014, 23(5): 2356-2368 doi: 10.1109/TIP.2014.2313227 [14] Wang D, Lu H C, Yang M H. Online object tracking with sparse prototypes. IEEE Transactions on Image Processing, 2013, 22(1): 314-325 doi: 10.1109/TIP.2012.2202677 [15] Zhang K H, Zhang L, Yang M H. Real-time compressive tracking. In: Proceedings of the 2012 European Conference on Computer Vision. Florence, Italy: Springer Berlin Heidelberg, 2012. 864-877 [16] 石武祯, 宁纪锋, 颜永丰.压缩感知跟踪中的特征选择与目标模型更新.中国图象图形学报, 2014, 19(6): 932-939 http://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB201406014.htmShi Wu-Zhen, Ning Ji-Feng, Yan Yong-Feng. Feature selection and target model updating in compressive tracking. Journal of Image and Graphics, 2014, 19(6): 932-939 http://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB201406014.htm [17] 毛征, 袁建建, 吴珍荣, 曲劲松, 李红岩.基于在线特征选择的实时压缩跟踪.光学精密工程, 2014, 22(3): 730-736 doi: 10.3788/OPE.Mao Zheng, Yuan Jian-Jian, Wu Zhen-Rong, Qu Jin-Song, Li Hong-Yan. Real-time compressive tracking based on online feature selection. Optics and Precision Engineering, 2014, 22(3): 730-736 doi: 10.3788/OPE. [18] 钟权, 周进, 吴钦章, 王辉, 雷涛.一种改进的实时压缩跟踪算法.光电工程, 2014, 41(4): 1-8 http://www.cnki.com.cn/Article/CJFDTOTAL-GDGC201404002.htmZhong Quan, Zhou Jin, Wu Qin-Zhang, Wang Hui, Lei Tao. An improved real-time compressive tracking. Opto-Electronic Engineering, 2014, 41(4): 1-8 http://www.cnki.com.cn/Article/CJFDTOTAL-GDGC201404002.htm [19] 罗会兰, 钟宝康, 孔繁胜.结合目标预测位置的压缩跟踪.中国图象图形学报, 2014, 19(6): 875-885 http://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB201406008.htmLuo Hui-Lan, Zhong Bao-Kang, Kong Fan-Sheng. Object tracking algorithm by combining the predicted target position with compressive tracking. Journal of Image and Graphics, 2014, 19(6): 875-885 http://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB201406008.htm [20] Zhang K H, Zhang L, Yang M H. Fast compressive tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(10): 2002-2015 doi: 10.1109/TPAMI.2014.2315808 [21] Comaniciu D, Ramesh V, Meer P. Real-time tracking of non-rigid objects using mean shift. In: Proceedings of the 2000 IEEE Conference on Computer Vision and Pattern Recognition. Hilton Head Island, USA: IEEE, 2000, 2: 142-149 [22] Wu Y, Lim J, Yang M H. Online object tracking: a benchmark. In: Proceedings of the 2013 IEEE Conference on Computer Vision and Pattern Recognition. Portland, USA: IEEE, 2013. 2411-2418 [23] Everingham M, Van Gool L, Williams C L I, Winn J, Zisserman A. The pascal visual object classes (VOC) challenge. International Journal of Computer Vision, 2010, 88(2): 303-338 doi: 10.1007/s11263-009-0275-4 期刊类型引用(1)
1. 尚婷，钱富才，刘磊，胡绍林. 一类混合不确定系统的对偶自适应控制. 应用科学学报. 2018(06): 1022-1030 . 
百度学术其他类型引用(4)
-
图(10) / 表(4)
计量
- 文章访问数: 2235
- HTML全文浏览量: 258
- PDF下载量: 856
- 被引次数: 5
