一种基于高斯混合模型的运动目标检测改进算法  

Improved Moving Object Detection Algorithm Based on Gaussian Mixture Model

作　　者：陈世文[1] 蔡念[1] 唐孝艳[1] 

机构地区：[1]广东工业大学信息工程学院,广东广州510006

出　　处：《现代电子技术》2010年第2期125-127,130,共4页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(30600121);广东省自然科学基金资助项目(07301038)

摘　　要：在运动目标检测方法的研究中提出一种基于高斯混合模型的运动目标检测的改进算法。首先利用颜色信息对背景建立高斯混合模型;其次在模型更新阶段,为了模型的自适应性和尽量逼近真实信号量,在传统学习率基础上提出一种加权思想,即对均值与方差分别给出一个不同的加权值。最后应用中值滤波及物体空间连通性进行后处理。实验结果表明,与传统高斯混合模型方法相比,改进的方法能更加有效地检测出运动目标,具有较好的鲁棒性。An improved method for detecting moving objects based on Gaussian mixture model for a video monitoring system is proposed in detection research. A Gaussian mixture model is established for the background based on color information. To keep adaptive characteristics and approximation to the real signal, a weighted method is provided based on the traditional learning rate in the model updating step, which gives different weights to the mean value and the variance value correspondingly. A median filter and object spatial connectivity are used to reduce noise. Experimental results indicate that compared with the old method,the proposed method can detect the moving objects more effectively,which is a robust method.

关 键 词：运动目标检测 高斯混合模型 加权值 鲁棒性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]