一种结合帧差法和混合高斯的运动检测算法  

A MOVING DETECTION METHOD COMBINING FRAME DIFFERENCE AND GAUSSIAN MIXTURE MODEL

作　　者：於正强[1] 潘赟[2] 宦若虹[3] 

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]浙江大学信息与电子工程学系,浙江杭州310027 [3]浙江工业大学计算机科学与技术学院,浙江杭州310023

出　　处：《计算机应用与软件》2015年第4期129-132,270,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61204030;61302129);浙江省自然科学基金项目(LQ12F04002;LY13F020030)

摘　　要：为了克服传统相邻帧差法和混合高斯模型的不足,提出一种结合改进的相邻帧差法和混合高斯模型的运动检测方法。对于相邻帧差法,增加了梯度比较和中值滤波,突出了运动目标边界和抗噪声干扰能力。通过结合相邻帧差法和混合高斯模型的结果,把图像分为4种区域,对每个区域内的混合高斯模型参数采用不同的更新策略,减少了运动目标对背景模型的干扰,对于物体在运动与静止之间转换的情况也具有很好的适应性。根据区域的划分情况和相邻帧差法的结果确定最终的运动区域。实验表明,既缓解了相邻帧差法产生的空洞问题,也消除背景物体突然转为运动物体后产生的"影子"问题。To overcome the drawbacks of the traditional adjacent frame difference method and Gaussian mixture model,we present a mov-ing object detection method which combines the improved adjacent frame difference and Gaussian mixture model.For the adjacent frame difference it adds gradient comparison and median filter,stands out the moving object boundary and enhances the ability of noise interference resistance.By combining the results of adjacent frame difference and Gaussian mixture model,we divide image into 4 regions.Different up-date strategies are adopted to the parameters of Gaussian mixture models in each region,this reduces the interference of moving object on background model and has a better adaptability on the situation of objects switching between motion and stillness.The moving region is even-tually determined according to the partition situation of regions and the result of adjacent frame.Experiment demonstrates that the combination of two methods mitigates the “hole”problem of the adjacent frame difference,and also eliminates the “shadow”problem when a background object suddenly switches to moving object.

关 键 词：运动目标检测 相邻帧差法 混合高斯模型 梯度比较 中值滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]