Journal of South China University of Technology(Natural Science Edition) ›› 2011, Vol. 39 ›› Issue (12): 63-69.

• Computer Science & Technology • Previous Articles     Next Articles

Vehicle Shadow Elimination Based on Multi-Point Pair Reference Model

Qi Qi-feng1,2  Wu Xin-sheng1,2  Deng Jun1   

  1. 1. School of Automation Science and Technology,South China University of Technology,Guangzhou 510640,Guangdong,China; 2. Engineering Research Center for Precision Electronic Manufacturing Equipment of the Ministry of Education,South China University of Technology,Guangzhou 510640,Guangdong,China
  • Received:2011-04-01 Revised:2011-09-08 Online:2011-12-25 Published:2011-11-04
  • Contact: 戚其丰(1976-) ，男，博士，讲师，主要从事数字图像处理、机器视觉等的研究． E-mail:qqifeng@ gmail.com
  • About author:戚其丰(1976-) ，男，博士，讲师，主要从事数字图像处理、机器视觉等的研究．
  • Supported by:

    国家自然科学基金资助项目( 60835001) ;华南理工大学中央高校基本科研业务费专项资金资助项目( 2009ZM0143)

PDF

720

Abstract:

In order to eliminate the shadow interference in a vehicle detection process,a stable multi-point pair reference model based on point pair properties,which is insensitive to the environment,is established in the global domain of the image. This model takes advantage of the hue and the luminance among the foreground,the background and the moving shadow in HSL color space and adopts the offline training method in a given background image set. By fully considering the color information of the whole image,the model reduces the error of background pixels,online
eliminates the impact of moving shadow on vehicle segmentation and strongly inhibits the moving background and illumination change in complex environments. Moreover,to simplify the computation of the proposed algorithm,background templates are introduced,which greatly reduces the number of operation pixels and improves the segmentation efficiency. Simulated results indicate that,as compared with the other vehicle shadow elimination methods,the proposed algorithm based on the multi-point pair reference model is more accurate and robust.

Key words: multi-point pair reference model, shadow elimination, vehicle segmentation, color space