基于改进的混合高斯模型的目标检测  

OBJECT DETECTION BASED ON IMPROVED GAUSSIAN MIXTURE MODEL

作　　者：郭伟[1] 刘鑫焱[1] 肖振久[1] 

机构地区：[1]辽宁工程技术大学软件工程学院,辽宁兴城125105

出　　处：《计算机应用与软件》2015年第2期148-150,166,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(61103199);北京市自然科学基金项目(4112052)

摘　　要：混合高斯模型已经广泛应用于背景建模中,但是结果受到噪音的干扰和光照突变的影响。为了解决这个问题,将Stauffer的混合高斯模型进行改进,当帧间差分判断出场景变化时,每个像素点的学习率会随着变化。由于边缘图像受到噪音干扰小,将这种改进的混合高斯模型也应用在边缘图像中,来提取边缘前景。边缘前景膨胀后,通过原图像的前景和边缘前景的与运算,得到最后的结果。实验结果表明,可以很好地去除噪音和解决光照突变的影响,提高了目标检测的效果,比传统方法更加有效。Gaussian mixture model has been widely applied in background modelling. But their outcomes are interfered by noises and impacted by sudden changes in illumination. To solve the problem, we improve the Gaussian mixture model suggested by Stauffer, when the scene variation is determined by the interframe difference, the learning rate in each pixel point will change accordingly. This improved Gaussian mixture model is also applied to edge images to extract the edge foreground images because the edge images are less sensitive to noise. After the dilatation in edge foreground, we get the final result through AND operation on the foreground of original images and the edge foreground. Experimental results indicate that the proposed method can well restrain noise and deal with the impact from sudden changes in illumination, improves the effect of object detection, and is more efficient than traditional methods.

关 键 词：混合高斯模型 边缘检测 目标检测 学习率 帧间差分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]