Journal of Computer Applications ›› 2017, Vol. 37 ›› Issue (9): 2610-2616.DOI: 10.11772/j.issn.1001-9081.2017.09.2610

Previous Articles     Next Articles

Video monitoring method of escalator entrance area based on Adaboost and codebook model

DU Qiliang, LI Haozheng, TIAN Lianfang   

  1. College of Automation Science and Engineering, South China University of Technology, Guangzhou Guangdong 510640, China
  • Received:2017-03-23 Revised:2017-05-17 Online:2017-09-10 Published:2017-09-13
  • Supported by:
    This work is partially supported by the Projects on the Integration of Industry, Education and Research of Guangzhou (201604010114), the Special Funds for Frontier and Key Technology Innovation of Guangdong (2016B090912001), the International Cooperation Projects of Science and Technology Information Bureau of Guangzhou (2012J5100001).

基于Adaboost和码本模型的手扶电梯出入口视频监控方法

杜启亮, 黎浩正, 田联房   

  1. 华南理工大学 自动化科学与工程学院, 广州 510640
  • 通讯作者: 黎浩正,466739850@qq.com
  • 作者简介:杜启亮(1980-),男,广东佛山人,副研究员,博士,主要研究方向:机器人、机器视觉;黎浩正(1994-),男,广东番禺人,硕士研究生,主要研究方向:计算机视觉、机器学习;田联房(1969-),男,山东济宁人,教授,博士,主要研究方向:模式识别、人工智能。
  • 基金资助:
    广州市产学研项目(201604010114);广东省前沿与关键技术创新专项资金资助项目(2016B090912001);广州市科信局国际合作项目(2012J5100001)。

Abstract: Aiming at the problem that the traditional video monitoring method can not divide the dense foreground objects accurately, a multi-target video monitoring method based on Adaboost and codebook model was proposed. Firstly, the Adaboost human head classifier was obtained by training, and the background model was established for the vertical elevator image by the codebook algorithm. The foreground image was extracted and heads were detected and tracked. After that, the pedestrian targets were removed to get the object targets, and the object targets were tracked. Finally, the movement of pedestrians and objects was monitored. The experimental results on 12 entrance area videos show that the method can track pedestrians and objects accurately and stably. It can accomplish the monitoring tasks of retrograde detection, passenger statistics, pedestrian congestion and object retention. With the processing speed of 36 frames per second, the tracking-accuracy rate is above 94% and the monitoring-accuracy rate is 95.8%. The proposed algorithm meets robustness, real-time and accuracy requirements of the intelligent video monitoring system.

Key words: Adaboost, background modeling, video monitoring, head detection, multi-target tracking

摘要: 针对传统视频监控方法无法对密集前景目标进行准确分割的问题,提出一种基于Adaboost和码本模型的多目标视频监控方法。首先,通过训练得到Adaboost人头分类器,利用码本算法为垂直拍摄的手扶电梯出入口图像建立背景模型,提取前景图像对其进行人头检测和跟踪;之后,剔除行人目标得到物件目标,对物件目标进行跟踪;最后,根据行人和物件的运动特征进行监控。对12段出入口视频序列的实验结果表明,监控方法能够准确稳定地跟踪行人和物件,完成逆行检测、客流统计、行人拥堵和物件滞留等监控任务,处理速度达到36帧/秒,目标跟踪准确率达到94%以上,行为监控准确率达到95.8%,满足智能视频监控系统鲁棒性、实时性和准确性的要求。

关键词: Adaboost, 背景建模, 视频监控, 人头检测, 多目标跟踪

CLC Number: