文章导读
总览 评价 邓琳 1, , 刘英豪 2, , 贾配洋 2, , 张雪波 1,* ( 1、 南开大学信息技术科学学院,天津 300071; 2、 南开大学信息技术科学学院,天津 300071; ) 摘要: 针对监控视频中的中、低密度人群,采用基于像素的方法对人群进行图像分析以得到前
邓琳1,, 刘英豪2,, 贾配洋2,, 张雪波1,*
(
1、南开大学信息技术科学学院,天津 300071; 2、南开大学信息技术科学学院,天津 300071; )
摘要:
针对监控视频中的中、低密度人群,采用基于像素的方法对人群进行图像分析以得到前景图像的总像素数和边缘图像的总像素数;针对监控视频中的高密度人群,采用纹理分析的方法从统计的角度,利用灰度共生矩阵提取出纹理特征,并进行归一化;再利用融合判别方法,根据前景像素占整个背景像素的比例将当前帧图像判定为中低密度或高密度;利用中低密度情况下前景图像的总像素数和边缘图像的总像素数与人群密度存在正相关的关系及高密度情况下纹理特征与人群密度之间的线性相关性,采用最小二乘曲线拟合的方法分别得到中、低密度及高密度下的人群密度曲线,实现了具有较高准确度的针对不同密度的人群的数目估计。
关键词:
视频监控;人群密度估计;像素法;灰度共生矩阵;最小二乘拟合
Deng Lin1,, Liu Yinghao2,, Jia Peiyang2,, Zhang Xuebo3,*
(
1、Information Science and Technology School,Nankai University, Tianjin 300071; 2、Information Science and Technology School,Nankai University,Tianjin 300071; 3、Information Science and Technology College,Nankai University,Tianjin 300071; )
Abstract:
In view of the surveillance video of the middle and low density of population, the method based on pixels for people to get the foreground image after image analysis of the total image pixels and edge pixels; For high density crowd in surveillance video, we adopt the method of texture analysis from the perspective of statistics, and based on the gray level co-occurrence matrix to extract the texture features, then normalize the results; Use assessment method of fusion to determine low or high density based on foreground pixels` proportion in the whole background pixels of the current frame image. Finally, because of the positive correlation among the foreground image pixels、edge pixels and population density, and the linear correlation between texture features and population density , we use the method of least squares respectively to get the crowd density curves in meddle、low and high density of population , and realized the estimation for different population density with higher accuracy.
Tag:
点此返回栏目查看更多>>>参考论文