摘要
针对海量互联网视频图片和监控视频中游行示威自动检测问题,提出了一种新的检测方法。该方法将游行示威检测问题分解为人群识别、横幅检测、人群密度估计以及联合判别4个步骤。在人群识别、横幅检测和人群密度估计阶段,均采用了具有较高准确度的深度学习算法;在联合判别阶段,利用邻域检测方法判断人群与横幅间关系。试验表明,该方法可对游行示威的视频和图片进行检测,并具有较好的召回率和准确率。
Aimed at the problem of the automatic demonstration detection in massive internet video image and surveillance video,a new detection method is proposed.The method divides the demonstration detection problem into four steps,including the crowd recognition,the banner detection,the crowd density estimation and the joint judge.In the phase of the crowd recognition,the banner detection and the crowd density estimation,the high accuracy deep learning algorithm is adopted.In the phase of joint judge,the neighborhood detection method is used to judge the relationship between the banner and the crowd.The experiment shows that this method can detect the video and image of demonstration,and also has the good recall rate and accuracy rate.
作者
丁頠洋
刘格
王贤哲
邵玮炜
DING Weiyang;LIU Ge;WANG Xianzhe;SHAO Weiwei(The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210007,China;The 2nd Military Representative Office in Nanjing,Department of Equipment Development,Nanjing 210007,China)
出处
《指挥信息系统与技术》
2018年第6期75-79,共5页
Command Information System and Technology
关键词
深度学习
人群识别
横幅检测
人群密度估计
联合判别
deep learning
crowd recognition
banner detection
crowd density estimation
joint judge
