摘要
对于一些公共场所,人数的统计是安全领域的重要问题。为了获得大视角的监控视频,以得到更多有用的视频信息,同时兼顾实时性和监控成本,本文使用鱼眼镜头对人群进行监控。为了使所有视场范围内景物能够在有限大小的成像面上成像,设计者人为的引入桶形畸变,而且由于实际场景中的景物通过鱼眼镜头后,会产生权重不等的像素点,本文针对鱼眼图像透视加权模型,采用加权特征提取的方法,提取鱼眼图像中人群目标的特征,包括加权面积、加权周长、加权KLT角点数以及加权边缘点数这4个特征,并使用多元线性回归法估计出了人群中的人数,取得了良好的效果。
In some public places,the statistics about the number of people is an important problem in security field.In order to receive large perspective,to get more useful video information,considering real-time and monitoring costs,this paper used a fisheye lens to monitor population. In order to make all the field of view imaging features image in the finite size of the surface,the designer introduced the barrel distortion artificially. At the same time,because the actual scenery through a fisheye lens,will produce the pixels with different weight,in the process of crowd counting based on fisheye video image,four features including weighted area,weighted edge density,weighted KLT corner number and weighted contour perimeter are extracted based on the perspective weight model of fisheye camera,and polynary linear regression combined with these features is applied for crowd counting in fisheye images.The experimental results are quite good.
出处
《电子器件》
CAS
北大核心
2014年第6期1111-1115,共5页
Chinese Journal of Electron Devices
关键词
鱼眼视频图像
人数估计
透视加权模型
多元线性回归
fisheye video image
counting
perspective weight model
polynary linear regression