改进的基于高斯混合模型的运动目标检测方法

Improved moving objects detection method based on Gaussian mixture model

针对固定摄像机的视频监控系统,提出了一种改进的基于混合高斯模型的运动目标检测方法。在模型学习方面,均值与方差采用了不同的学习率,其中均值更新采用自适应的学习率,方差的学习率取固定值;引入权值均值概念,然后结合权值进行像素点的前景和背景分类;利用了背景图像消除阴影。实验结果表明,改进的方法与传统方法相比具有更好的学习能力,能提高在繁忙场景中,大而慢的运动目标检测的正确率。

This paper proposed an improved moving objects detection method based on Gaussian mixture model in the case of focusing on a video monitoring system with a static camera. First, for updating the parameters （ mean and variance） of the Gaussian models, the learning rates of mean and variance were different： for mean, the learning rate was adaptive, while for variance, the learning rate was fixed; Second, The notion of Mean Of the Weight （MOW） was introduced, which had a big contribution for differentiating background points from foreground points. Third, Shadow was detected and removed with the help of background image. Experimental results show that the proposed method possesses better ability of learning and higher efficiency of detecting large and slow objects in busy environments .