摘要
针对传统的混合高斯模型对动态背景敏感、缓变目标检测不准确等问题,提出了一种基于时空分布的混合高斯建模改进方法。该方法的基本思想是混合高斯背景基于时间分布信息建模的同时,通过随机数生成方法对邻域进行采样,完成像素空间分布的背景建模;同时利用像素历史统计信息和决策融合机制的前景检测方法,实现对静止目标判定以及前景运动目标更精确的提取。最后,将此算法与其他前景检测方法进行对比实验,表明了该算法对动态背景鲁棒性强、缓变目标检测准确的结论。
Considering the traditional GMM is sensitive to dynamic environment,has low detection rate for the slow moving target,this paper proposed a modified GMM algorithm,which used spatial information to compensate time information. It used the random number generation method for sampling the neighborhood to complete background modeling based on pixel spatial distribution during the time distribution based on Gaussian mixture background modeling. Meanwhile,it utilized the foreground detection algorithm,which applied pixel＇s history statistic information and decision fusion mechanism,to get a more exact judgment on static and moving target. Finally,compared with other foreground detection algorithms,it shows that GMM has better robustness and more exact detection for the slow moving target.
出处
《计算机应用研究》
CSCD
北大核心
2015年第5期1546-1548,1553,共4页
Application Research of Computers
基金
广西自然科学基金资助项目(2013GXNSFBA019278)
广西高校科学预研基金资助项目(2013YB032)
关键词
背景建模
空间信息
混合高斯模型
动态背景
前景检测方法
background modeling
spatial information
GMM
dynamic background
foreground detection algorithm
