基于纹理特征的混合高斯背景建模算法研究

Research on Gaussian Mixture Background Modeling Algorithm Based on Texture Feature

在智能交通系统中,运动目标的检测是一个基本而又关键的问题。而传统高斯混合模型能较好地检测出运动目标,但由于其没有考虑像素的局部特征,使得运动目标区域的错误检测率有所增加。为了更好地在高速交通视频中检测出完整且准确的运动目标前景区域,文中在子空间的思想基础上,提出一种基于像素局部纹理特征的高斯混合模型改进算法,即以像素周围5*5图像块的均值、标准差、最大值、最小值和当前像素值5个特征作为局部纹理特征,建立高斯混合背景模型,进行运动目标检测。经过大量实验,结果表明该算法能更准确、完整地检测出运动目标并具有很好的环境适应性,特别是在运动目标区域与相应的背景区域颜色较为相似时,运动目标检测效果改善较为明显。

Moving target detection in the intelligent transportation system is a fundamentaland key issue. Traditional Gaussian mixture model can better detect moving targets, but without considering the local characteristics of pixels, resulting in the error detection rate increases of moving target. To solve these problem, on the basis of idea of subspace, an improved Gaussian mixture model algorithm based on local texture features for pixel is put forward. It uses the average, standard deviation, maximum, minimum, and current pixel values around pixel 5 ＊ 5 image block as local texture features, and establishes Gaussian mixture background model for moving object detection. After extensive comparison of experimental results, it shows that the algorithm can more accurately and completely detect moving targets and has good environmental adaptability. When the color of moving target area is similar with corresponding background area,the detection results improved is obvious.