改进的自适应高斯混合模型运动目标检测算法  

Moving target detection algorithm for improved adaptive Gaussian mixture model

作　　者：唐洪良 黄颖[2] 黄淮[2] 杨成顺[2] 黄宵宁[2] 

机构地区：[1]国网浙江杭州市余杭区供电公司,浙江杭州311100 [2]南京工程学院电力工程学院,江苏南京211167

出　　处：《现代电子技术》2017年第11期65-67,共3页Modern Electronics Technique

基　　金：南京工程学院引进人才科研启动基金(YKJ201412)

摘　　要：传统的高斯混合模型学习率和分布数取值固定,不能精确地描述变换的背景,同时存在数据冗余等问题,针对这些不足,进行了以下三个方面的改进。在模型初始化阶段,针对不同的环境设定各异的初始分布数;根据环境变化快慢程度,动态调整学习率的取值;不断更新高斯分布,删除不满足要求的模型,创建新的分布。实验结果表明,改进的自适应高斯混合模型较传统高斯混合模型,显著提高了运动物体检测的准确性。Since the traditional Gaussian mixture model has fixed learning rate and distribution number,can′t describe the transformation background accurately,and exists the data redundancy,the Gaussian mixture model was improved in the following three aspects. In the stage of the model initialization,the various initial distribution numbers are set according to different environments. The value of learning rate is adjusted dynamically according to the speed of the environmental change. The Gaussian distribution is updated continuously to delete the model that can′t satisfy the requirement,and create the new distribution. The experimental results show that,in comparison with the traditional Gaussian mixture model,this adaptive Gaussian mixture model can improve the detection accuracy of moving object significantly.

关 键 词：高斯混合模型 运动物体检测 高斯分布 学习率取值 

分 类 号：TN912.34[电子电信—通信与信息系统]