首页 | 本学科首页   官方微博 | 高级检索  
     

混合高斯模型运动检测算法优化
引用本文:胥 欣,江登表,李 勃,陈启美.混合高斯模型运动检测算法优化[J].计算机应用研究,2013,30(7):2190-2194.
作者姓名:胥 欣  江登表  李 勃  陈启美
作者单位:南京大学 电子科学与工程学院, 南京 210046
基金项目:国家自然科学基金资助项目(61105015); 交通运输部科技项目(2012-364-641-209); 江苏省科技支撑计划资助项目(BE2011747, BE2009667); 江苏省自然科学基金资助项目(BK2011511); 江苏省环境监测科研基金资助项目(0917)
摘    要:针对经典混合高斯算法对非平稳场景过于敏感的问题, 提出了混合高斯模型运动检测算法的优化方法。在检测算法流程的匹配高斯模态选择、模型更新和背景显示上分别作了优化:综合考虑模态权重与模态自身匹配度, 选择匹配高斯模态; 统一初始化与检测过程中的模型更新, 即使视频检测的背景变化较大, 系统也能较快地建立理想背景模型; 综合考虑背景模型各模态分布, 清楚反映了背景模型的具体分布。实验结果表明, 与经典混合高斯算法相比, 优化算法在克服背景扰动、降低误检率上表现良好, 有效提高了混合高斯算法对场景变化的适应性。优化算法在实际工程应用中效果良好。

关 键 词:智能视频监控    运动检测    背景建模    混合高斯模型

Optimization of Gaussian mixture model motion detection algorithm
XU Xin,JIANG Deng-biao,LI Bo,CHEN Qi-mei.Optimization of Gaussian mixture model motion detection algorithm[J].Application Research of Computers,2013,30(7):2190-2194.
Authors:XU Xin  JIANG Deng-biao  LI Bo  CHEN Qi-mei
Affiliation:School of Electronic Science & Engineering, Nanjing University, Nanjing 210046, China
Abstract:Considering the sensitivity of Gaussian mixture model in non-stationary scenes, the proposed method improved the Gaussian mixture model algorithm, especially in three ways. The method combined the weights and models themselves while choosing the best matched Gaussian mode, and used the same update method at system initializing and motion detecting process, creating an ideal background model as quickly as possible, even if the background had significant changes, and considered the background model distribution of each mode, while displaying the background, making the show of background reflect the specific distributions of the background model clearly. Compared with the classical Gaussian mixture algorithm, the proposed method does well in overcoming the background disturbances and reducing the false detection rate. The proposed method also has a good performance in engineering applications.
Keywords:intelligent video surveillance  moving object detection  background modeling  Gaussian mixture model
点击此处可从《计算机应用研究》浏览原始摘要信息
点击此处可从《计算机应用研究》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号