基于改进高斯混合模型的运动目标检测方法  

Moving Target Detection Method Based on Gaussian Mixture Model

作　　者：程全[1,2] 马军勇[3,4] 

机构地区：[1]郑州大学信息工程学院,郑州450001 [2]周口师范学院物理与电子工程学院,周口466001 [3]光电控制技术重点实验室,洛阳471009 [4]中航工业洛阳电光设备研究所,洛阳471009

出　　处：《计算机科学》2014年第7期318-321,共4页Computer Science

基　　金：河南省科技厅重点科技攻关资助项目:基于移动互联网和物联网的家庭保健中心研究(132102210101)资助

摘　　要：基于高斯混合模型,提出了一种自适应的运动目标检测算法。首先,根据各像素点的像素值的集中程度,自适应地选择高斯分布的个数对背景模型进行学习与更新,再通过背景差分获取差分图像;其次,在对图像二值化的过程中,提出了一种改进的自动调整阈值的方法,用以对差分图像的像素进行分类后分别进行阈值化分割,这样就能得到前景目标;接着采用形态学重构的方法对阴影进行有效消除,从而使前景目标分割的效果得到有效的提高。实验证明,该方法具有较好的鲁棒性和检测效果,同时也具有较好的自适应性,特别是在检测目标本身灰度变化比较大等特殊情况下,更能体现出本算法的优越性。Based on Gaussian mixture model, this paper put forward an adaptive moving target detection algorithm. First of all,according to the intensity of pixel values of each pixel, the number of Gaussian distribution is adaptively selected to learn and update background model, again get difference image by background subtraction. Second, in the process of image binarization, an improved method of automatic threshold is proposed to separately classifly difference image pixels after threshold segmentation, so future goal can be obtained. Then the method of morphological reconstruction is adopt- ed to effectively eliminate the shadow, significantly improving prospect target segmentation effect. Experiment proves that the method has good robustness and detection effect, as well as good adaptability. Especially when the grayscale change of detection target itself is large, the superiority of the algorithm is more obvious.

关 键 词：混合模型 图像 形态学 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]