Abstract: This paper presents an improved background generation algorithm based on sequence image. On the foundation of building the Gaussian mixture model, the algorithm categories the Gaussian distributions which correspond to a pixel into reliable distributions and unreliable distributions that can evolve into reliable distributions by accumulating pixel values. When a new image comes up, the algorithm will create a new distribution, update an existed distribution or remove an obsolete distribution according to a specified pixel of the current image, and update the background image at the same time, to get a real-time background without moving targets. A quantitative analysis is given to the experimental data, result proves the adaptation and stability of the proposed algorithm.

Key words: background generation, Gaussian distributions, sequence image, moving target

摘要： 提出一种基于序列图像的改进的多高斯分布背景生成算法。该算法在用多高斯分布背景中每个像素建模的基础上，把每个像素对应的若干个高斯分布分成可靠分布和不可靠分布，依据序列图像中每幅当前图像的像素值创建新分布、更新已有分布和删除过时分布来更新背景模型，并实时生成不含运动目标的背景图像。算法定量分析了实验数据，结果证明，该算法具有良好的自适应能力和稳定性。

关键词: 背景生成, 多高斯分布, 序列图像, 运动目标

CLC Number: 

• TP391