在线加权多示例学习实时目标跟踪

陈东成; 朱明; 高文; 孙宏海; 杨文波

doi:10.3788/OPE.20142206.1661

您当前的位置：

首页 >

文章列表页 >

在线加权多示例学习实时目标跟踪

信息科学 | 更新时间：2020-08-13

- 在线加权多示例学习实时目标跟踪
- Real-time object tracking via online weighted multiple instance learning
- 光学精密工程 2014年22卷第6期页码：1661-1667
- 作者机构：
  
  1. 中国科学院长春光学精密机械与物理研究所,吉林长春,中国,130033
  2. 中国科学院大学北京,中国,100049
  3. 中国科学院长春光学精密机械与物理研究所中国科学院航空光学成像与测量重点实验室,吉林长春,130033
- 作者简介：
- 基金信息：
  
  中国科学院航空光学成像与测量重点实验室开放基金面上项目（No.Y2HC1SR121）()
- DOI：10.3788/OPE.20142206.1661
  中图分类号： TP391
- 收稿日期：2013-04-22，
  
  修回日期：2013-06-05，
  
  纸质出版日期：2014-06-25
- 稿件说明：
移动端阅览
陈东成, 朱明, 高文等. 在线加权多示例学习实时目标跟踪[J]. 光学精密工程, 2014,22(6): 1661-1667

CHEN Dong-cheng, ZHU Ming, GAO Wen etc. Real-time object tracking via online weighted multiple instance learning[J]. Editorial Office of Optics and Precision Engineering, 2014,22(6): 1661-1667
陈东成, 朱明, 高文等. 在线加权多示例学习实时目标跟踪[J]. 光学精密工程, 2014,22(6): 1661-1667 DOI： 10.3788/OPE.20142206.1661.

CHEN Dong-cheng, ZHU Ming, GAO Wen etc. Real-time object tracking via online weighted multiple instance learning[J]. Editorial Office of Optics and Precision Engineering, 2014,22(6): 1661-1667 DOI： 10.3788/OPE.20142206.1661.

摘要

由于原始多示例学习（MIL）跟踪的分类效果和实时性较差，提出了一种加权在线多示例学习跟踪算法。首先，根据所选定目标位置分别采集目标和背景样本集，通过对所采集样本集特征的在线学习生成弱分类器集；然后，用计算样本集对数似然函数的最大值的方法从弱分类器集中选择

个最优的弱分类器，给每个弱分类器赋不同的权值，生成一个强分类器；最后，在新的一帧中抽取目标和背景样本，用生成的强分类器对待分类的目标和背景进行分类；分类结果映射成概率值，概率最大样本的位置就是所要跟踪目标的位置。对不同视频序列的测试结果表明，该跟踪算法的跟踪正确率达93%，目标大小为43 pixel×36 pixel时处理帧率约为25 frame/s。与原始多示例学习跟踪算法相比，本算法的实时性提高了67%。

Abstract

A weighted Multiple Instance Learning(MIL) tracking method was proposed to improve the precision and real-time quality of online MIL tracking algorithm. First

target samples and background samples around a selected target were collected. Weak classifiers were generated by online learning the features of collected samples. In order to get

best weak classifiers

the maximum of samples' log-likelihood was calculated. Every weak classifier was weighted differently and

weak classifiers were combined into a strong classifier. Finally

new unclassified samples were picked from the newly formed frame. The obtained strong classifier was used to separate the target and background. The classifying results were mapped into probabilities and the location of the sample with the largest probability was the target location wanted. Experiments on variant videos show that the accurate rate of the proposed algorithm is 93% and the average frame rate is 25 frame/s when the object size is 43 pixel×36 pixel. Compared with the original MILtracking algorithm

the real-time quality of proposed method increases by 67%.

关键词

Keywords

references

颜佳,吴敏渊. 遮挡环境下采用在线Boosting的目标跟踪[J]. 光学精密工程,2012,20(2):439-446. YAN J, WU M Y. On-line boosting based target tracking under occlusion [J]. Opt. Precision Eng., 2012,20(2):439-446.(in Chinese)

ZHOU Q H, LU H CH, YANG M H. Online multiple support instance tracking[J]. IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, 2011,545-552.

王田,刘伟宁,韩广良,等. 基于改进MeanShift的目标跟踪算法[J]. 液晶与显示,2012,27(3):396-400. WANG T,LIU W N,HAN G L. Target tracking algorithm based on improved Meanshift [J]. Chinese Journal of Liquid Crystals and Displays, 2012, 27(3):396-400.(in Chinese)

李英,李静宇,徐正平. 结合SURF与聚类分析方法实现运动目标的快速跟踪[J]. 液晶与显示,2011,26(4):544-550. LI Y, LI J Y, XU ZH P. Moving target fast tracking using SURF and cluster analysis method [J]. Chinese Journal of Liquid Crystals and Displays, 2011, 26(4):544-550.(in Chinese)

龚俊亮,何昕,魏仲慧,等. 采用改进辅助粒子滤波的红外多目标跟踪[J]. 光学精密工程,2012,20(2):413-421. GONG J L, HE X, WEI ZH H, et al.. Multiple infrared target tracking using improved auxiliary particle filter[J]. Opt. Precision Eng., 2012,20(2):413-421. (in Chinese)

沈凌云,郎百和,朱明. 一种基于人工神经网络的人脸识别方法[J]. 液晶与显示,2011,26(6):836-840. SHEN L Y, LANG B H, ZHU M. Face recognition method based on artificial neural network[J]. Chinese Journal of Liquid Crystals and Displays, 2011, 26(6):836-840.(in Chinese)

DIETTERICH T G, LATHROP R H, LOZANO-PREZ T. Solving the multiple-instance problem with axis-parallel rectangles[J]. Artificial Intelligence, 1997, 89(1-2): 31-71.

VIOLA P, PLATT J C, ZHANG C. Multiple instance boosting for object detection[J]. Proc. Neural Information Processing systems, 2005,1417-1426.

BABENKO B, YANG M, BELONGIE S, Robust object tracking with online multiple instance learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33:1619-1632.

ZEISL B, LEISTNER C, SAFFARI A, et al.. On-line semi-supervised multiple-instance learning boosting. 2010 IEEE Conference on Computer Vision and Pattern Recognition,2010:1879.

PAPAGEORGIOU C P, OREN M, POGGIO T. A general framework for object detection [C].Sixth International Conference on Computer Vision, 1998:555-562.

OZA N. Online Ensemble Learning [D]. Berkeley:University of California, 2001.

MASON L, BAXTER J, BARTLETT P, et al.. Functional Gradient Techniques for Combining Hypotheses [M]. Cambridge: MIT Press, 2000,221-247.

KALAL Z, MATAS J, MILKOLAJCZYK K. Online learning of robust object detectors during unstable tracking [C]. OLCV, 2009.

KALAL Z, MILKOLAJCZYK K, MATAS J. Tracking-learning-detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012,1409-1422.

浏览量

223

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

数据驱动的多模复合制导信息融合及其试验验证

基于非机动单站的目标运动状态无偏估计

地基大口径望远镜动态目标跟踪中压电式快速反射镜迟滞效应的补偿

基于级联滑模控制的高精度光电跟踪与捕获

引入样本删除机制的TLD粒子群目标跟踪