一种基于混合高斯的双空间自适应背景建模方法

A double-subspace adaptive background modeling method based on Gaussian mixture model

为解决运动目标缓慢运动或暂时停止以及场景突变问题,受人类获取知识过程启发,提出一种基于混合高斯的双空间自适应背景建模方法,即采用当前混合高斯模型空间和记忆空间(用于存储曾经的背景模型)对场景进行自适应建模。两个空间模型更新采用不同的学习率:在当前混合高斯模型空间,学习率根据高斯分布对场景的贡献程度进行自适应更新,以解决运动目标缓慢运动或暂时停止问题;记忆空间存储曾经的背景模型,以提高算法对背景突变的适应性,故采用固定学习率进行更新。试验结果表明了所提方法的优越性。

In order to tackle problems that the moving object slows down or stops for a while, and the background changes suddenly when segregating the foreground from background ;＂ inspired by the human learning process, a double-subspace a- daptive background modeling method based on Gaussian mixture model（ GMM ） was proposed. A memory space is introduced into the traditional GMM-based background modeling for storing the past background models. The learning rates for updating the distributions in the two spaces are different. In GMM space, the learning rate is updated with the contribution of the dis- tribution to the scene, which aims to handle problems that the object moves slowly or stops temporarily. While in the memory space, a fixed learning rate is used in order to improve the adaptability to sudden background changes. The experimental re- sults demonstrate the superiority of the proposed method.