Abstract: Gaussian mixture modeling has become the standard method for background modeling of the implementation of background subtraction in video sequences, because of its ability to track multiple pixel value distributions in complex variational scenes. This paper analyzes the theory framework underlying the method, and illuminates the two aspects to improve the method：one is updating of model parameters, the other is the classifying tactics. Based on the proposed algorithm, the survey discusses controlling of learning rate, adaptive number of Gaussian models, algorithm evaluation and algorithm initializations. These analysis and discussions expect to provide new ideas and directions for the further research.

Key words: background subtraction, Gaussian Mixture Model（GMM）, Expectation Maximization algorithm（EM）, model evaluation

摘要： 混合高斯模型能够拟合像素颜色值分布、跟踪复杂的场景变化，基于它的算法已经成为对视频序列实施背景减法时的一个标准背景建模方法。分析了GMM算法的理论框架，提出了算法改进的两个方面：模型参数更新和BG/FG分类决策。在综述各种已有的算法的基础上，从学习因子控制、模态个数调节、算法评价以及算法初始化等几个方面展开分析。这些分析结果将为后续研究提供思路和方向。

关键词: 背景减法, 高斯混合模型, 最大期望算法, 模型评价