基于模糊神经网络大场景人群密度估计方法.docVIP

基于模糊神经网络大场景人群密度估计方法 　　摘 要:提出了一种估计大场景下密集人群密度的方法。该方法根据人类视觉的模糊性原理,认为用模糊集来划分人群密度范围比用确定的方法更符合人眼视知觉的认知方式,利用统计的方法确定灰度共生矩阵各指标对于各个密度类别的隶属函数;设计基于误差反向传播训练算法(BP)的模糊神经网络,计算样本模式对于各个密度类别的隶属度,并根据人群密度变化的时域连续性原理对人群密度范围进行合理估计。实验表明该方法提高了估计精度。 　　关键词:人群密度估计; 模糊神经网络; 灰度共生矩阵; 智能视频监控 　　中图分类号:TP391.4文献标志码:A 　　文章编号:1001-3695(2010)03-0989-03 　　doi:10.3969/j.issn.1001??3695.2010.03.049 　　 　　 　　 　　Crowd density estimation of wide scene based on fuzzy neural network 　　 　　 　　TANG Qing??1, WANG Zhi-yan??1, YAN He-ping??2, XU Xiao-wei????3a,3b?? 　　 　　(1.School of Computer Science Engineering, South China University of Technology, Guangzhou 510006, China; 2. No.75771 Unit of PLA, Guangzhou 510540, China; 3 a.Shool of Information Science Technology,b.Key Laboratory of Digital Life, Sun Yat?Msen University, Guangzhou 510275,China) 　　 　　Abstract:This paper proposed a crowd density estimation method based on a fuzzy neural network. According to the fuzzy phenomena used by human vision, it was more accordant with cognitive style of human vision by using fuzzy sets to describe crowd density range than other definite measures. Defined functions of membership degree of indicators of the grey level dependence matrix (GLDM) by statistical method and designed a fuzzy neural network based on the error back propagation (BP) training algorithm to calculate the membership degree of the input pattern, which could be explained reasonably according to the temporal continuity of variety of the crowd density. Experimental results show this method performs well and improves the accuracy of estimation. 　　Key words:crowd density estimation; fuzzy neural network; grey level dependence matrix; intelligent video surveillance 　　 　　 　　 　　0 引言 　　随着人类文明的发展和社会的进步,对众多的公共场所, 如广场、车站等地方进行人数统计是公共安全领域的重要课题[1,2]。传统方法是人工依据经验来确定监控视频画面中聚集人群的数量、拥挤程度等,费时费力且缺乏客观性。因此,应用计算机视觉技术实现场景中的人群智能监控统计成为当前的研究热点。 　　现有工作主要分为两类:第一类方法是人体目标特征分析法[3,4]。首先检测、跟踪视频序列图像中的人体特征,识别是否人体目标,然后进行统计。这类方法在室内的小场景下效果较好,适用于人数比较稀疏情况。然而,对于大场景情况下的公共场所而言,普遍