摘要
由于像素统计方法在提取高密度人群特征时,可能会导致在计算感兴趣区域(ROI)中的人数时出现较大的误差,因此提出归一化前景目标像素提取人群特征,并采用支持向量机(SVM)对ROI中的人群密度进行估计。首先利用混合高斯模型消除背景,并用Otsu算法提取人群目标,然后进行归一化前景目标像素的人群特征提取,最后利用支持向量机DAG算法实现人群密度分类,并与人工神经网络方法、基于像素的和基于纹理的方法进行了对比。实验结果表明正确检测率可达到95%。
When extracting high density crowd features based on pixel statistical methods,it may lead to bigger errors in estimating the number of crowd within the region of interest( ROI). Therefore,we presented a technique to extract crowd features based on normalised foreground target pixels,and adopted support vector machine( SVM) to estimate the crowd density in ROI. First,we used the mixture Gaussians model to remove the image background and the Otsu algorithm to extract crowd targets,then we employed the normalised foreground target pixels method to extract crowd features. Finally,we used the DAG algorithm of SVM to achieve the classification of the crowd density.Moreover,we compared the presented technique with artificial neural network approach and the methods based on pixels and textures.Experimental results showed that the rate of true detection could be up to 95%.
出处
《计算机应用与软件》
CSCD
2016年第4期212-214,296,共4页
Computer Applications and Software
关键词
人群密度估计
归一化前景目标
人群特征
支持向量机
Crowd density estimation
Normalised foreground target
Crowd feature
Support vector machine