高斯核密度估计的背景建模运动目标检测  

Moving Target Detection based on Gauss Kernel Density Estimation Background Modeling

作　　者：燕莎[1] 潘永[2] 

机构地区：[1]西安理工大学,陕西西安710082 [2]西安热工研究院有限公司,陕西西安710054

出　　处：《探测与控制学报》2017年第1期76-79,83,共5页Journal of Detection & Control

摘　　要：针对视频图像中光照的渐变和突变等引起的动态背景和图像前景中运动目标(物体)存在阴影等问题,提出了高斯核密度估计的背景建模运动目标检测方法。实验结果表明,该方法采用了非参数密度估计理论,像素特征的概率分布不需要预先假设,同时估计出来的像素特征的概率密度函数更符合真实的背景像素的概率分布,能够处理多样性的动态背景场景,具有较强的背景适应能力,能够准确地提取运动目标,从而有效检测出视频图像中的运动目标。The modeling for motion background due to illumination changes and sudden changes or shadow is the key in the motion target （object） detection field.A method based on kernel density estimation theory for background image reconstruction was presented in this paper.Experimental results showed that,by using non-parametric density estimation,the probability distribution of pixel features do not need to assume in advance,calculating the probability density function by the above method,and it was much close to real probability distribution of background image pixels.The proposed method could be effectively build background for the variety cases of background changes in videos,the algorithm was robust and could accurately extract motion target （object） with the background difference method.

关 键 词：背景建模 目标检测 混合高斯模型 背景减法 核密度估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]