计算机应用 ›› 2011, Vol. 31 ›› Issue (07): 1831-1834.DOI: 10.3724/SP.J.1087.2011.01831

• 图形图像技术 • 上一篇    下一篇

复杂条件下高斯混合模型的自适应背景更新

李明之,马志强,单勇,张晓燕   

  1. 空军工程大学 电讯工程学院,西安 710077
  • 收稿日期:2010-12-22 修回日期:2011-02-22 发布日期:2011-07-01 出版日期:2011-07-01
  • 通讯作者: 李明之
  • 作者简介:李明之(1983-), 男, 湖南醴陵人, 硕士研究生, 主要研究方向:视频图像处理;马志强(1968-),男,山东乳山人,副教授,硕士, 主要研究方向:图像处理、多媒体通信;单勇(1976-),男,河南睢县人,讲师,博士,主要研究方向:图像分析理解、信息融合;张晓燕(1970-),女,陕西西安人,副教授,博士,主要研究方向:图像与视频处理。

Adaptive background update based on Gaussian mixture model under complex condition

Ming-zhi LI,Zhi-qiang MA,Yong SHAN,Xiao-yan ZHANG   

  1. Institute of Telecommunication Engineering, Air Force Engineering University, Xi’an Shaanxi 710077,China
  • Received:2010-12-22 Revised:2011-02-22 Online:2011-07-01 Published:2011-07-01
  • Contact: Ming-zhi LI

摘要: 针对高斯混合模型背景更新中面临的光照突变和目标与背景相互转化的问题,提出一种分情况分区域的背景自适应更新算法。首先根据当前检测目标的面积大小判别是否发生光照突变情况,采取针对性更新策略,对于未发生光照突变情况再分背景区域和目标区域分别进行背景自适应更新。其中,重点讨论了目标区域的背景更新问题,提出根据目标尺寸、运动速度和匹配次数等特征参数来调整目标区域的背景更新速率。仿真结果表明,该算法在保证了目标检测完整性的同时,提高了模型对背景变化的适应能力。

关键词: 高斯混合模型, 背景更新, 运动目标检测

Abstract: In view of the background update problems based on Gaussian Mixture Model (GMM), such as sudden illumination change, mutual transformation between target and background, a new background auto-adapted update algorithm was proposed in this paper. First, the algorithm distinguished whether the sudden change of illumination occurred according to the current tested target size, and took pertinent updating measures. If there is not sudden change of illumination, the backgrouds of backgroud region and target region were updated seperately. The update of the target region was mainly discussed, and the method that adjusted the background update rate of the target region according to the targets characteristic parameters, such as size, velocity of movement and match times was proposed. The simulation results show that the algorithm not only guarantees the integrity of target detection, but also improves the adaptation of model to background changes.

Key words: Gaussian Mixture Model (GMM), background updating, moving objects detection