计算机科学 ›› 2015, Vol. 42 ›› Issue (2): 296-300.doi: 10.11896/j.issn.1002-137X.2015.02.063

• 图形图像与模式识别 • 上一篇    下一篇

基于方向矢量的多特征融合粒子滤波人体跟踪算法研究

张蕾,宫宁生,李金   

  1. 南京工业大学电子与信息工程学院 南京211816,南京工业大学电子与信息工程学院 南京211816,南京工业大学电子与信息工程学院 南京211816
  • 出版日期:2018-11-14 发布日期:2018-11-14
  • 基金资助:
    本文受国家重点基础研究发展计划(973计划)(2005CB321901),软件开发环境国家重点实验室开放课题(BUAA-SKLSDE-09KF-03)资助

Research of Human Tracking Algorithm through Multi Feature Fusion Particle Filter Based on Direction Vector

ZHANG Lei, GONG Ning-sheng and LI Jin   

  • Online:2018-11-14 Published:2018-11-14

摘要: 针对传统的多特征融合粒子滤波跟踪算法计算量大、不利于实时性、人群拥挤遮挡时容易出现跟踪匹配错误等情况,提出了基于方向矢量的多特征融合粒子滤波跟踪算法。该算法首先将人体颜色特征与轮廓特征进行乘性融合和加性融合后相加并加上两者的不确定性的乘积,以便能够根据两种特征的实际贡献率来调节各自在跟踪过程中所占的权重比例,从而提高了跟踪的准确性;其次结合方向矢量,根据先前的跟踪信息来预测运动物体可能运动的范围从而减少了粒子迭代计算量;最后通过动态调节窗口将合并的人体进行分离处理。实验证明,本方法能够在复杂情况下对人体进行实时准确的跟踪。

关键词: 粒子滤波,多特征融合,人体跟踪,不确定性

Abstract: Traditional multi feature fusion particle filter algorithm has huge amount of computation,which is not conducive to timeliness.Meanwhile,the algorithm often generates error for tracking and matching.To solve these problem better,this paper adopted the multi feature fusion particle filter algorithm founded in direction vector.Firstly,the algorithm adds the value of multiplicative fusion and additive fusion through the body color feature and contour feature.In addition,on the basis of the uncertainty multiplication of two features,the respective weights proportion in the tracking process is adjusted according to the actual contribution rate of two kinds of features.Moreover,the probable range of the movement of objects can be predicted to reduce the amount of calculation of particle iteration by combining the direction vector.At last,the merge human’s body can be separated by adjusting the window automatically and precisely.The test shows our method can realize the human tracking accurately in the complex environment.

Key words: Particle filter,Multi feature fusion,Human tracking,Uncertainty

[1] 王玉茹,刘家峰.基于多区域联合粒子滤波的人体运动跟踪[J].自动化学报,2009,35(11):1387-1393
[2] 田纲,胡瑞敏,王中元.一种基于运动矢量分析的Mean-shift目标跟踪算[J].中国图象图形学报,2010,15(1):85-90
[3] Shi Yong-gang,Karl W C.Real-time Tracking Using Level Sets[J].Pattern Recognition,2005,23(12):34-41
[4] 蔺海峰,马宇峰,宋涛.基于SIFT特征目标跟踪算法研究[J].自动化学报,2010,36(8):1024-1028
[5] 王永忠,梁彦,赵春晖,等.基于多特征自适应融合的核跟踪方法[J].自动化学报,2008,34(4):393-399
[6] 张焱,张志龙,沈振康.一种融入运动特性的显著性特征提取方法[J].国防科技大学学报,2008,30(3):109-115
[7] 陈祖爵,陈潇君,何鸿.基于改进的混合高斯模型的运动目标检测[J].中国图象图形学报,2007,12(9):1586-1589
[8] 徐梦溪,刘超颖,黄陈蓉.基于多级信息融合的视频目标根据[J].南京工程学院学报:自然科学版,2010,8(2):1-9
[9] Nickels K,Hutchinson S.Estimating uncertainty in ssd-basedfeature tracking[J].Image Vision Computer,2002,0(1):47-58
[10] 赵凤,范九伦.一种结合二维Ostu法和模糊熵的图像分割方法[J].计算机应用,2007,24(6):189-191

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!