摘要
公共场景监控下的人群密度估计已经是公共安全管理中的一个重要环节。为了提高对视频监控中人群密度估计的实时性和准确率,提出一种改进的混合高斯背景建模进行前景提取,并用大小随目标边缘可变的矩形框对人群目标进行圈定以代替传统的边缘像素数和前景像素数;通过最小二乘法拟合估计人数和实际人数的线性关系,使用平均相对误差和平均绝对误差进行定量对比分析。实验结果表明:与基于边缘像素统计和阈值分割像素统计的算法相比,该算法能够直接统计出有效人数,较为准确的估计出视频图像中的人群数目,且误差最低。
Crowd density estimation under the public scene monitoring is an important part of public security management,in order to improve real-time and accuracy of crowd density estimation in the video monitoring,put forward an improved gaussian mixture background modeling to extract the foreground and use the size with the target edge variable rectangular box on the crowd to identify areas instead of the traditional edge pixels and foreground pixels,fitting by the least squares method to estimate the number and the actual number of linear relationship,using the average relative error and mean absolute error to compare the analysis quantitatively. The experimental results show that: compared with statistics based on edge pixels and compared with threshold segmentation algorithm of pixel statistics,the algorithm can directly statistic the number of crowd effectively,estimating the number of people in video images more accurately,and the error is minimum.
出处
《科学技术与工程》
北大核心
2017年第17期266-271,共6页
Science Technology and Engineering
关键词
视频图像
混合高斯背景建模
人群密度估计
矩形框
video image
gaussian mixture background modeling
crowd density estimation
rectangular box