深圳大学学报理工版

刘磊,陈泽虹,张勇,等.深度图像下基于头部多特征的人数统计算法[J].深圳大学学报理工版,2017,34(No.6(551-660)):584-590.[doi:10.3724/SP.J.1249.2017.06584]
Liu Lei,Chen Zehong,Zhang Yong,et al.A people counting algorithm based on multi-feature of head region in depth images[J].Journal of Shenzhen University Science and Engineering,2017,34(No.6(551-660)):584-590.[doi:10.3724/SP.J.1249.2017.06584]

点击复制

深度图像下基于头部多特征的人数统计算法

刘磊¹,陈泽虹²,张勇¹,赵东宁^2,3

1)深圳大学ATR国防科技重点实验室, 广东深圳518060; 2)深圳大学信息工程学院, 广东深圳518060; 3)哈尔滨工业大学(深圳)计算机科学与技术学院,广东深圳 518055

关键词：数字图像处理; 深度图像; 区域生长; Kalman滤波; 多特征; 人数统计

A people counting algorithm based on multi-feature of head region in depth images

Liu Lei¹, Chen Zehong², Zhang Yong¹, and Zhao Dongning^2,3

Liu Lei¹, Chen Zehong², Zhang Yong¹, and Zhao Dongning^2,31)ATR Key Laboratory of National Defense Technology, Shenzhen University, Shenzhen 518060, Guangdong Province, P.R.China2)College of Information Engineering, Shenzhen University, Shenzhen 518060, Guangdong Province, P.R.China3)School of Computer Science and Technology, Harbin Institute of Technology(Shenzhen), Shenzhen 518055, Guangdong Province, P.R.China

Keywords： digital image processing; depth image; region growing; Kalman filter; multi-feature; people counting

DOI: 10.3724/SP.J.1249.2017.06584

备注

摘要

全文

图/表

参考文献

在现实生活中,因人流量过大而引发的安全事故不胜枚举.为了防止此类事故的发生,可通过视频监控的方式统计人数,及时对行人进行限流和分流.提出一种有效的人数统计算法.该算法采用深度摄像机作为视频采集源,通过分析和提取深度图像下头部的4个特征,实现行人头部检测,并依靠Kalman滤波技术实现对头部目标的跟踪,进而达到人数统计的目的 .该算法对行人的不同发型具有一定适应性,同时对轻微遮挡和多人环境下的头部检测均有良好效果.该算法人数统计平均准确率达到88.6%.

In daily life, a great number of security accidents are caused by the excessive flow of people. In order to prevent the occurrence of such accidents, we propose an efficient algorithm to count the number of people by using video monitors and limit the flow of people in time. The algorithm uses the depth camera as a video capture device and realizes the detection of people's heads by analyzing and extracting the four features of heads in depth image. The method uses Kalman filter technology to track the head and achieves the purpose of counting statistics. The proposed algorithm can effectively solve the head detection problem of complex scenes, such as hairstyle diversity and head part-occlusion. The average accuracy of the proposed algorithm reaches about 88.6%.