Please wait a minute...
浙江大学学报(工学版)
J4  2013, Vol. 47 Issue (6): 977-983    DOI: 10.3785/j.issn.1008-973X.2013.06.007
计算机技术     
半监督Hough Forest跟踪算法
林亦宁, 韦巍, 戴渊明
浙江大学 电气工程学院, 浙江 杭州 310027
Semi-supervised Hough Forest tracking method
LIN Yi-ning, WEI Wei, DAI Yuan-ming
College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China
 全文: PDF  HTML
摘要：

针对基于检测的目标跟踪算子存在着目标表观描述的准确性和对跟踪过程的目标表观变化的适应性之间的矛盾,提出一种较为柔性的半监督学习方法：以Hough Forests为基本学习框架,用离线学习的分类器作为先验跟踪算子，并引入基于跟踪过程中用检测得到的object-specific信息和空间一致性信息的随机标签分布,用来生成半监督学习所需要的在线数据样本,对跟踪算子进行修正|分类器的检测和跟踪环节与在线的Hough Forests分类器应用环节类似,但采用了类似粒子滤波的随机采样方式对检测和更新环节进行加速|用匀速运动模型对目标运动进行建模,使跟踪过程中的目标位置预测更加准确.算法在标准跟踪数据集合i-Lids和TUD-campus上与当前流行的目标跟踪算法在线Boosting算法、MILB和Hough Forests跟踪算法进行了比较实验.实验证明：监督Hough Forests分类器提供了解决分类器对目标表观表述的准确性和自适应性矛盾的一种有效机制|整个跟踪算法能够使目标跟踪过程更加快速、鲁棒与准确.
Abstract:

State-of-art tracking-by-detection methods have somewhat conflict between accuracy of object appearance description and adaptivity to object appearance changing in tracking process. Focused on this difficult problem, a flexible semi-supervised learning method based on Hough Forests classifier was proposed. First, an off-line classifier as prior tracker for object appearance description was used. Then, a random label distribution, which is based on the spatial consistency in context and the object-specific information detected in tracking, was introduced to generate on-line examples for updating semi-supervised on-line classifier. Detection and updating phases were carried out in a similar way as normal on-line Hough Forests classifier goes, however, a particle-filtering-kind random sampling scheme was implemented in detection and updating phases to accelerate tracking. Finally, a uniform motion model was employed to help object position prediction more accurate. Experiments were carried out on standard visual tracking databases such as i-Lids and TUD-campus, and the results were compared with those of the popular tracking algorithms such as on-line Boosting, MILB and Hough Forest tracking. Experimental results show that this semi-supervised Hough Forests classifier provides an optimal fashion to efficiently solve the accuracy/adaptivity conflict of classifier|meanwhile, the proposed tracking algorithm can significantly improves the tracking performance in tracking rate, robustness and precision.
出版日期: 2013-11-22
:  TP 181  
基金资助:

国家“863”高技术研究发展计划资助项目(84861)；国家自然科学基金资助项目(60704030)；中央高校基本科研业务费专项资金资助项目.
通讯作者: 韦巍,男，教授，博导.     E-mail: wwei@cee.zju.edu.cn
作者简介: 林亦宁(1984—)，男，博士生，从事计算机视觉、机器学习方向研究.E-mail: lynphoenix@gmail.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章　　

引用本文:

林亦宁, 韦巍, 戴渊明. 半监督Hough Forest跟踪算法[J]. J4, 2013, 47(6): 977-983.

LIN Yi-ning, WEI Wei, DAI Yuan-ming. Semi-supervised Hough Forest tracking method. J4, 2013, 47(6): 977-983.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2013.06.007    或    http://www.zjujournals.com/eng/CN/Y2013/V47/I6/977

［1］ YILMAZ A, JAVED O, SHAH M. Object tracking: A survey［J］. ACM Computing Surveys, 2006, 38(4)：13.

［2］ OZA N C. Online bagging and boosting［C］∥ Systems, man and cybernetics, 2005 IEEE international conference on. \
[S.L.\]: IEEE, 2005, 3: 2340-2345.

［3］ FFARI A, LEISTNER C, SANTNER J, et al. On-line random forests［C］∥ ICCV 2009. Japan: IEEE, 2009: 1393-1400.

［4］ GALL J, LEMPITSKY V. Class-specific Hough forests for object detection［C］∥ CVPR 2009. USA: IEEE, 2009: 1022-1029.

［5］ CHAPELLE O, SCHOLKOPF B, ZIEN A. Semi-Supervised Learning［M］. Cambridge: The MIT Press, 2006.

［6］ ZHU X. Semi-supervised learning literature survey［R］. Madison, USA University of Wisconsin-Madison, 2005.

［7］ BREIMAN L. Random forests［J］. Machine Learning, 2001, 45(1): 532.

［8］ OKODA R. Discriminative generalized hough transform for object detection［C］∥ ICCV 2009. Japan: IEEE, 2009: 2000-2005.

［9］ MAJI S, MALIK J. Object detection using a max-margin hough transform［C］∥ CVPR 2009. Miami, USA: IEEE, 2009: 1038-1045.

［10］ LEISTNER C, SAFFARI A, SANTNER J, et al. Semi-supervised random forests［C］∥ICCV 2009. Japan: IEEE, 2009: 506-513.

［11］ Home Office Scientific Development Branch. Imagery library for intelligent detection systems i-lids［EB/OL］.http:∥www.elec.qmul.ac.uk/staffinfo/andrea/avss2007_d.html.

［12］ ANDRILUKA M, ROTH S, SCHIELE B. People-tracking-by-detection and people-detection- by- tracking［C］∥CVPR 2008. Alaska, USA: IEEE, 2009: 18.

［13］ GRABNER H, GRABNER M, BISCHOF H. Real-time tracking via on-line boosting［C］∥ BMVC 2006. Britain: IEEE, 2006: 47-56.

［14］ BABENKO B, YANG M H, BELONGIE S. Visual tracking with online multiple instance learning ［C］∥ CVPR 2009. Miami, USA: IEEE, 2009: 983-990.

［15］ EVERINGHAM M, GOOL L V, WILLIAMS, et al. The Pascal visual object classes(VOC) challenge ［J］. International Journal of Computer Vision, 2010, 88 (2):303-338.
[1] 李侃,黄文雄,黄忠华. 基于支持向量机的多传感器探测目标分类方法[J]. J4, 2013, 47(1): 15-22.
[2] 王洪波, 赵光宙, 齐冬莲, 卢达. 一类支持向量机的快速增量学习方法[J]. J4, 2012, 46(7): 1327-1332.
[3] 艾解清, 高济, 彭艳斌, 郑志军. 基于直推式支持向量机的协商决策模型[J]. J4, 2012, 46(6): 967-973.
[4] 潘俊, 孔繁胜, 王瑞琴. 局部敏感判别直推学习机[J]. J4, 2012, 46(6): 987-994.
[5] 金卓军, 钱徽, 朱淼良. 基于倾向性分析的轨迹评测技术[J]. J4, 2011, 45(10): 1732-1737.
[6] 顾弘, 赵光宙. 广义局部图像距离函数下的图像分类与识别[J]. J4, 2011, 45(4): 596-601.
[7] 罗建宏,陈德钊. 兼顾正确率和差异性的自适应集成算法及应用[J]. J4, 2011, 45(3): 557-562.
[8] 商秀芹, 卢建刚, 孙优贤. 基于遗传规划的铁矿烧结终点2级预测模型[J]. J4, 2010, 44(7): 1266-1269.
[9] 徐磊, 赵光宙, 顾弘. 成对耦合分类器的多球体预处理方法[J]. J4, 2010, 44(2): 237-242.