Abstract: In some gradient and repetitive motion modeling scenes, traditional Gaussian mixture background modeling has a good effect. But the algorithm needs a large amount of computation and storage space, and it can neither fit for complex background or background with sudden changes. To solve these problems, a new clustering background modeling based on human memory model is proposed. Combined with human memory model, the algorithm sets up a cluster model for each pixel, and each cluster can be adaptively created, updated and deleted according to background changes. The algorithm makes accurate judgments according to human ultra-short-term memory, short-term memory and long-term memory, and the moving target detection results can meet the judgment of the human sensory organs.

Key words: background modeling, memory model, cluster, moving target

摘要： 传统的混合高斯背景建模可对存在渐变及重复性运动的场景进行建模，但其运算过程需要足够的计算量和存储空间，不适应在复杂背景下的背景建模，也不能解决场景中存在的突变。针对这些问题，提出了一种基于记忆模型的聚类算法，算法为每个像素点设置一个聚类模型，每个聚类可根据背景的变化结合人类记忆模型自适应的创建、更新和删除。该算法通过人类瞬时记忆、短时记忆和长时记忆做出准确判断，运动目标检测结果更能符合人类感觉器官的判断。

关键词: 背景建模, 记忆模型, 聚类, 运动目标