一种基于GMM的目标检测改进算法  

An Improved Target Detection Algorithm Based on GMM

作　　者：江汉红[1] 熊玮佳[2] 李庆[2] 

机构地区：[1]海军工程大学电气与信息工程学院,武汉430033 [2]武汉理工大学信息工程学院,武汉430070

出　　处：《武汉理工大学学报》2013年第3期132-135,共4页Journal of Wuhan University of Technology

基　　金：国家自然科学基金(50977090);湖北省自然科学基金(2010CDB01503)

摘　　要：针对经典的GMM(Gaussian Mixture Model)背景差分法中模型参数更新策略容易使高斯分布的方差迅速趋近于一个极小值的缺陷,提出了一种新的模型参数学习机制,使得高斯模型的方差平稳收敛,避免了陷入方差过小的恶性循环,最终在一定程度上减少目标误判的发生。结果表明,提出的改进算法在光照突变情况下,能够克服背景的扰动,目标误判的可能性也大大降低,与经典的GMM检测算法相比,改进后的算法在检测的精度和抗干扰性上有了明显的提高。The update strategy of the model parameters in traditional background subtraction of classic Gaussian Mix- ture Model can easily make the variance of the model approach a very small value rapidly. In order to overcome this prob- lem, this paper put forward a new learning mechanism of the model parameters, so that the Gaussian model got a conver- gent variance, and this mechanism can avoid the vicious cycle of its falling into a too small one. As a result, the misjudg- ment of the target could be suppressed at a certain extent. The experimental results indicate that the improved algorithm can overcome the disturbance of the background while the light mutation happens accidentally, and the possibility of the misjudgment of targets reduced efficiently. Compared with the traditional GMM, the accuracy and interference immunity of the detection has been enhanced obviously by using the improved algorithm.

关 键 词：图像处理 高斯分布 模型参数学习机制 目标误判 抗干扰性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]