Computer Engineering and Applications ›› 2015, Vol. 51 ›› Issue (5): 228-233.

Previous Articles     Next Articles

Triple background modeling based detection of violative behaviors against city appearance

XU Li1, WU Baozhou1, SHI Bin1, YANG Jie2   

  1. 1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China
    2.School of Civil Engineering and Architecture, Zhejiang University of Science and Technology, Hangzhou 310023, China
  • Online:2015-03-01 Published:2015-04-08

基于三背景建模的市容违章检测

许  力1,吴宝洲1,史  彬1,杨  洁2   

  1. 1.浙江大学 电气工程学院,杭州 310027
    2.浙江科技学院 建筑工程学院,杭州 310023

Abstract: City appearance violative behavior management is an important part of city management based on video. The system introduces a triple background modeling with feedback modification method aiming at the time retention characteri-stics of the violations of garbage hoarding and illegal road occupying. Triple backgrounds are long backgrounds, short backgrounds and standard backgrounds modeled by different update rates and modeling methods respectively. The long background is then modified by the feedback of detection result to exclude the stationary foreground. Statistical diagram is used to form feedback modification masks and illegal alarm areas. A method based on geometric characteristics and dispersion is put forward to detect and remove stationary person, and it is proposed to improve the further accuracy. Experimental results show that the algorithm can recognize violative behaviors efficiently, enhance the robustness of illegal stationary substances detection, reduce the computational complexity of the algorithm, and improve the detection accuracy further.

Key words: city appearance violation, triple background, feedback modification, statistical diagram, dispersion

摘要: 市容违章行为的检测是基于视频的城市管理的重要组成部分。针对垃圾囤积、占道经营等市容违章行为的时间滞留特点,提出一种带反馈修正机制的三背景建模方法。三背景包括长背景、短背景和标准背景,分别采用不同的更新速率和建模策略,通过对长背景的反馈修正,可防止可疑违章物更新其中;采用统计图形成反馈修正掩膜和可疑违章区域,并基于几何特征和离散度在可疑违章区域中检测并剔除静止人像。实验结果表明,该方法能有效地识别违章行为,具有较低的计算复杂度和较强的鲁棒性。

关键词: 市容违章, 三背景, 反馈修正, 统计图, 离散度