基于帧间特征和连通域检测的人数统计  

People Counting Based on Inter-frame Features and Connected Domain Detection

作　　者：朱家新[1] 郭德全[2] 兰时勇[1,3] 

机构地区：[1]四川大学视觉合成图形图像技术国防重点学科实验室,四川成都610065 [2]成都信息工程大学控制工程学院,四川成都610225 [3]四川大学计算机学院,四川成都610065

出　　处：《计算机技术与发展》2016年第2期101-104,109,共5页Computer Technology and Development

基　　金：国家"863"高技术发展计划项目(2013AA013802)

摘　　要：为了更好地管理机场、火车站及汽车站等人流密集的固定场所,文中提出一种简单、高效的人员流量统计算法。首先,将视频序列中的图像转化为灰度图像,利用帧差法提取前景图像并将其二值化;然后,利用上下边缘点匹配的连通域检测算法对前景图像中的连通域进行提取,利用人头的形状信息和灰度信息对连通域进行筛选;最后,通过对人头连通域的跟踪实现人数统计。在提出的人头跟踪计数算法中,利用人头连通域在视频帧中的信息和容器对人头特征进行跟踪计数。实验结果表明,该算法具有良好的人头跟踪和计数效果。In order to manage airports,railway stations and bus stations and other cro ded places fixed with a better way,a simple and efficient personnel traffic statistics algorithm was proposed. First,the video images was converted into grayscale images,and then the foreground image was extracted using the frame difference method and was converted into binary image. Second,connected domain was detected based on matching between the points of up- contour and dow n- contour,and then head features were extracted using shape and gray information of head. Finally,people was counted by tracking head connected domain. In this head tracking and counting algorithm presented,use the basic information and container of head connected domain in the video image for tracking and counting. Experimental results show that the algorithm has a good head tracking and counting results.

关 键 词：人头特征 跟踪 计数 连通域 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]