Abstract: This paper analyzes the background modeling mechanism using Gaussian mixture models, and then discusses the stability/plasticity dilemma in parameters estimation and update of GMM background model. After that, the importance of Gaussian component’s variance to motion segmentation is pointed out. To solve the slow convergence problem of Gaussian mean and variance update formula given by Stauffer, a new updating strategy is proposed, which weighs the model adaptability and motion segmentation accuracy. Experimental results show that the proposed algorithm improves the accuracy of modal learning and speed of variance convergence.

Key words: Gaussian Mixture Models（GMM）, background model, motion segmentation, parameter estimation

摘要： 在分析高斯混合背景模型建模机理的基础上，研究了模型参数估计及更新中模型结构稳定性和可塑性两难问题，指出高斯分量方差估计对运动分割的重要性。针对Stauffer算法中高斯分量均值和方差更新公式收敛过慢问题，提出了兼顾适应性和运动分割准确性的均值和方差更新策略。实验结果表明该方法在模态学习的准确性和方差收敛速度方面比原有方法有较大提高。

关键词: 高斯混合模型, 背景模型, 运动分割, 参数估计