基于多摄像头的会场上座率实时统计方法  

Real-Time Statistical Method of Venue Attendance Based on Multiple Cameras

作　　者：钱晨[1] 鲁帅[1] 冯瑞[1] 

机构地区：[1]复旦大学计算机科学技术学院,媒体计算研究所

出　　处：《微型电脑应用》2014年第5期14-17,共4页Microcomputer Applications

基　　金：科技部支撑计划(2013BAH09F01);上海市科委科技创新行动计划(12511501602)

摘　　要：会场中座位分布密集,摄像头采集的画面中座位经常受到遮挡,严重影响了上座率统计的准确性。针对这种情况,提出了一种基于多摄像头的会场上座率统计方法。在座位区域的正面和顶部安装摄像头,在两种视角的画面中运用背景差分算法筛选出发生状态变化的座位,有效减低算法复杂度,做到实时计算,然后对座位图像提取HOG特征,利用支持向量机(SVM)进行分类,最后融合两种视角的分类结果,减少遮挡带来的影响,统计得出会场上座率。实际应用结果表明,该方法在保证会场上座率统计实时性基础上,具有很好的鲁棒性和准确性。Because the distribution of seats in venue is dense,occlusion is very common in camera videos,which affects the accuracy of the attendance statistics.This paper proposes a real-time statistical method of venue attendance based on multiple cameras to this situation.Installed cameras in the front and top of seats firstly,using background subtraction algorithm in videos captured by them to decrease time cost by filtering out unchanged seats,which realizes the real-time detection method,and then detect seat status in two view angles by HOG feature and SVM.At last,fusing the results of HOG feature classification in front and top view angles to overcome the influence of occlusion.Application results have proved that this real-time method is robust and effective.

关 键 词：会场上座率 多摄像头 实时统计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]