您的位置:山东大学 -> 科技期刊社 -> 《山东大学学报(工学版)》

山东大学学报(工学版) ›› 2013, Vol. 43 ›› Issue (4): 39-45.

• 机器学习与数据挖掘 • 上一篇    下一篇

基于视频流的复杂场景公车人头对象计数研究

张国华,谭晓阳*,陈松灿   

  1. 南京航空航天大学计算机科学与技术学院, 江苏 南京 210016
  • 收稿日期:2013-05-14 出版日期:2013-08-20 发布日期:2013-05-14
  • 通讯作者: 谭晓阳(1972- ),男,重庆人,教授,博导,主要研究方向为机器学习、模式识别、计算机视觉. E-mail: x.tan@nuaa.edu.cn
  • 作者简介:张国华(1987- )男,江苏泰州人, 硕士,主要研究方向为模式识别与人工能. E-mail:domyselfzhang@qq.com.
  • 基金资助:

    国家自然科学基金资助项目(61073112, 61035003);江苏省自然科学基金资助项目(BK2012793)

Counting heads from bus video streams under uncontrolled conditions

ZHANG Guo-hua, TAN Xiao-yang*, CHEN Song-can   

  1. College of Computer Science & Technology, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, China
  • Received:2013-05-14 Online:2013-08-20 Published:2013-05-14

摘要:

以安装于公车内顶棚的单目摄像为研究对象,首先结合CENTRIST(census transform histogram)及LBP(local binary patterns)特征描述子对输入图像进行快速人头检测,然后结合视频流中包含的目标运动信息、被检测对象的外观信息、以及上车时间先验信息等来得到准确的人头数目。通过在现实场景下采集的真实公车视频流数据上的实验结果表明,该方法以接近实时的速度工作,在保证准确率的同时,能够有效减少复杂场景下人头计数中的误检率及漏检率。

关键词: 公车, LBP, 人头检测, 视频流分析, CENTRIST

Abstract:

A novel method for head counting was presented based on the video streams coming from the single-camera installed on the roof of buses. The key idea of this method was based on the principal of count-by-detection method while taking account other sources of information available. First, a fast and effective head detection method was proposed using multi-modal feature sets, including CENTRIST and LBP. Second, a decision for a valid head count was made by fusing the information from head motion and the prior knowledge of the length of time needed for an event of interested. Extensive experiments on a series of video streams collected under real-life scenarios demonstrated the feasibility and effectiveness of the proposed method.

Key words: video analysis, LBP, head detection, bus, CENTRIST

中图分类号: 

  • TP301
[1] 赵军伟1,侯清涛2,李金屏3,彭勃4. 基于数学形态学和HSI颜色空间的人头检测[J]. 山东大学学报(工学版), 2013, 43(2): 6-10.
[2] 严云洋1,2,唐岩岩2,刘以安2,张天翼3. 使用多尺度LBP特征和SVM的火焰识别算法[J]. 山东大学学报(工学版), 2012, 42(5): 47-52.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!