摘要： 为有效提取复杂场景中的运动前景，提出基于贝叶斯理论的Dirichlet共轭先验，结合在线最大似然估计(Online EM)改进基于高斯混合模型参数的背景减除算法。改进算法避免了陷入局部最值，在线自适应地调整高斯个数，并生成运动全景图，实验结果表明，该算法能有效提高前景检测率。

关键词: 背景减除, Dirichlet共轭先验, 在线最大似然估计, 贝叶斯理论, 高斯混合模型, 全景图

Abstract: With the challenge of extracting moving foreground objects from dynamic background, this paper introduces an improved background subtraction algorithm based on Gaussian Mixture Model(GMM) by using Dirichlet conjuagate prior and Online EM in Bayes framework. It avoids converging to a local maximum of the log-likelihood function, selects the numbers of Gaussian adaptively and outperforms the panorama. Experimental results demonstrate that the improved algorithm can increase the detection rate for foreground effectively.

Key words: background subtraction, Dirichlet conjuagate prior, Online EM, Bayes theory, Gaussian Mixture Model(GMM), panorama

中图分类号: 

• TP391.4