一种改进的混合高斯学习自适应背景建模算法

An Improved Adaptive Background Modeling Algorithm Based on Gaussian Mixture Learning

针对混合高斯学习模型计算复杂度高,实时响应系统应用困难等问题,提出了一种改进的背景建模算法,首先利用帧差法进行预处理,选择出帧间变化区域,然后对变化区应用混合高斯学习模型进行采样计算,完成视频背景建模。由于混合高斯学习模型融合了增量最大期望分类学习方法,自动选择学习率参数具有更好的收敛速度和背景估计精度;同时通过帧差法预处理降低了算法的计算量。实验表明,该算法在保证收敛稳定性和背景建模精度的情况下,提高了背景分割的响应速度。

For the high computational complexity of Gaussian Mixture Learning,hard application problems of real - time response system and other issues, this paper presents an improved scheme for modeling video background. In the completion of the video background modeling process, firstly the interchanging area is selected according to the frame difference method, and then the sampling calculation based on Gaussian Mixture Learning is applied to this selected area. Since the Gaussian mixture model combining improved Expectation Maximization classification learning methods,automatically parameter learning rate has a better convergence speed and estimation accuracy,and also the algorithm computation is reduced. The experimental result shows that the algorithm raises the response speed for background segmentation, without reducing the convergence stability and modeling accuracy.