计算机科学 ›› 2013, Vol. 40 ›› Issue (Z6): 199-203.

• 模式识别 • 上一篇    下一篇

基于HOG和Haar特征的行人追踪算法研究

陆星家,陈志荣,尹天鹤,杨帆   

  1. 宁波工程学院理学院 宁波315211;宁波工程学院理学院 宁波315211;宁波工程学院理学院 宁波315211;宁波工程学院理学院 宁波315211
  • 出版日期:2018-11-16 发布日期:2018-11-16
  • 基金资助:
    本文受国家自然科学基金项目(40901241),浙江省自然科学基金项目(Y5090377),浙江省教育厅基金项目(Y201225208),宁波市自然科学基金项目(2012A610020)资助

Research of Pedestrian Tracking Based on HOG Feature and Haar Feature

LU Xing-jia,CHEN Zhi-rong,YIN Tian-he and YANG Fan   

  • Online:2018-11-16 Published:2018-11-16

摘要: 行人在真实场景的检测和追踪是多目标检测和追踪研究中的一个重要问题,尤其是在真实的三维场景中的多行人之间的遮挡、拥挤以及背景的变化对多目标检测和追踪研究造成了严重的挑战。在多目标检测中利用了Haar特征、HOG特征,在行人正面向相机运动时,采用Haar特征检测器检测人脸,并结合Haar运动模型完成行人的检测,当行人侧向运动时,采用HOG特征,利用层次-部分模型进行行人的检测和追踪,在完成行人的检测之后,利用最大权重独立集合算法完成帧间目标的关联。通过对 ETH、TUD以及本地样本库的检测和追踪结果表明,采用Haar特征和HOG特征的检测算法对于行人的正面和侧面都具有较高的检测准确率、精确度。

关键词: Haar-Like特征,HOG特征,层次-部分模型,Haar运动模型,最大权重独立集

Abstract: The detection and tracking of pedestrians in the real scene is one of the most concerned questions in the multi-people detection and tracking,especially under the condition that the occlusion,congestion and varied backgrounds of people have greatly challenged the research of multi-people detection and tracking.Haar and HOG features are applied in this research.The Haar-Like feature detector is used to detect the face when people come face to face with the camera,while the Part-Based Template is used to detect and track people in people’s sideway movement.After that,the maximum weight independent set is used to accomplish the relationship between the frames.The result of the detection and tracking from the ETH,TUD and local sample library attests that the hybrid algorithm has much higher accuracy and precision in detecting pedestrian’s front and side face.

Key words: Haar-like feature,HOG feature,Part-based model,Haar motion model,Maximum weight independent set

[1] Xiao Jiang-jian,et al.Vehicle Detection and Tracking in Wide Field-of-View Aerial Video [C]∥IEEE Conf.on Computer Vision and Pattern Recognition.2010:13-18
[2] Cox I J,A review of statistical data association techniques for motion correspondence[J].International Journal of Computer Vision,1993,10(1):53-66
[3] Okuma K,Taleghani A,Freitas N D,et al.A boosted particle filter:Multitarget detection and tracking [C]∥European Confe-rence on Computer Vision.2004:28-39
[4] Brendel W,Amer M,Todorovic S.Multiobject tracking as maximum weight independent set [C]∥IEEE Conf.on Computer Vision and Pattern Recognition.2011:1273-1280
[5] Andriluka M,Roth S,Schiele B.People tracking by detectionand people detection by tracking [C]∥IEEE Conf.on Computer Vision and Pattern Recognition.2008
[6] Leibe B,Schindler K,Gool L V.Coupled detection and trajectory estimation for multi-object tracking [C]∥IEEE International Conference on Computer Vision.2007
[7] 程有龙,李斌,张文聪,等,融合先验知识的自适应行人跟踪算法[J].模式识别与人工智能,2009,22(5):704-708
[8] 李娟,邵春福,杨励雅,等.基于Kalman 滤波的行人跟踪方法研究[J].交通运输系统工程与信息,2009,9(6):148-153
[9] Felzenszwalb P F,Girshick R B,McAllester D.Discriminatively trained deformable part models [C]∥IEEE Conf.on Computer Vision and Pattern Recognition.2009
[10] Viola P,Jones M J,Snow D.Detecting Pedestrians Using Patterns of Motion and Appearance[J].International Journal of Computer Vision,2005,63(2):153-161
[11] Pavan M,Pelillo M,Dominant sets and pairwise clustering [J].PAMI,2007,29(1):167-172
[12] Ge J,Luo Y,Tei G.Real-time pedestrian detection and tracking at nighttime for driver-assistance systems [J].Intelligent Transportation Systems,IEEE Transactions on,2009,10(2):283-298
[13] Geismann P,Schneider G.A two-staged approach to vision-based pedestrian recognition using Haar and HOG features [C]∥Intelligent Vehicles Symposium,2008IEEE.IEEE,2008:554-559

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!