• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 图形与图像 • 上一篇    下一篇

基于改进混合高斯模型和图形句柄的异常车辆检测

刘艳萍1,崔彤1,周长兵2,李小翠2,刘甜1   

  1. (1.河北工业大学电子信息工程学院,天津 300401;2.中国地质大学(北京)信息工程学院,北京 100083)
  • 收稿日期:2019-06-28 修回日期:2019-08-29 出版日期:2020-02-25 发布日期:2019-02-25
  • 基金资助:

    河北省自然科学基金(E2016202341);河北省高等学校科学技术研究项目(BJ2014013)

Abnormal vehicle detection based on improved
 mixed Gaussian model and graphic handle

LIU Yan-ping1,CUI Tong1,ZHOU Zhang-bing2,LI Xiao-cui2,LIU Tian1   

  1. (1.School of Electronic Information Engineering,Hebei University of Technology,Tianjin 300401;
    2.School of Information Engineering,China University of Geosciences(Beijing),Beijing 100083,China)
     
  • Received:2019-06-28 Revised:2019-08-29 Online:2020-02-25 Published:2019-02-25

摘要:

由于交通安全隐患在当下的生活中造成的不良影响越发严重,所以在步行街、校园等禁止车辆行驶的场景中,对异常车辆的检测具有一定的现实意义。针对
利用混合高斯建立背景模型时易出现重影和空洞问题,提出了一种基于SSIM结构相似性的混合高斯建模的异常车辆检测,采用SSIM计算2幅图像像素点间的相似度,在高斯建模后进行二次背景建模,同时引入了指数函数来优化高斯建模过程中的权值更新过程,提高了更新速度。采用图形句柄函数优化连通域方法对前景区域进行异常车辆检测,能够检测出异常车辆且标注框更加贴近车辆形状。
对580幅由视频分割得到的图像的实验结果表明,检测率可以达到90.3%。
 

关键词: 混合高斯建模;SSIM, 图形句柄函数;异常车辆检测

Abstract:

Because the traffic safety hazards are becoming more serious in the current life, it has certain practical significance to detect abnormal vehicles in scenes such as pedestrian streets and campuses where vehicles are prohibited from traveling. Aiming at the problem of ghost and cavity in the background model built by mixed Gaussian, an abnormal vehicle detection based on mixed Gaussian modeling of SSIM structural similarity is proposed. The similarity between two image pixels is calculated by SSIM. The secondary background modeling is carried out after the Gaussian modeling, and the exponential function is introduced to optimize the weight update process in the Gaussian modeling process, which improves the update speed. The graphical handle function is used to optimize the connected domain method to detect the abnormal vehicle in the foreground area, it is possible to detect the abnormal vehicle and the labeling frame is closer to the shape of the vehicle. Experimental results  of 580 images segmented by video show that the detection rate can reach 90.3%.

 

Key words: Gaussian mixture modeling;SSIM, graphic handle function;abnormal vehicle detection