基于自适应学习率的背景建模方法

Background Modeling Approach Based on Self-adaptive Learning Rate

针对高斯混合模型中均值和方差的学习,提出基于自适应学习率的背景建模方法。统计每个像素模型被匹配的次数,在线更新学习率。在初始化背景时,分配一个全局的学习率,采用传统高斯混合模型的学习方式;在更新背景时,为每个像素分配一个学习率,采用自适应的学习方式。实验结果表明,该方法与传统高斯混合背景模型相比,有较好的学习能力与稳定性,能提高运动目标检测的正确率。

This paper proposes a background modeling approach based on self-adaptive learning rat aiming at the update of the learning rate about Gaussian mixture model. The initial background is established using the traditional Gaussian mixture model with a global learning rate. The self-adaptive learning rate is used for each pixel according to the number of matching when the background is updated. Experimental results show that compared with moving object detection approach based on conventional Gaussian mixture model, it has a desirable stability and learning ability.