基于GoogLeNet的智能录播系统中站立人脸的检测与定位  

Detection and Location of Standing Faces in Intelligent Recording and Broadcasting System Based on GoogLeNet

作　　者：衣柳成 魏伟波 刘小芳 YI Liu-cheng;WEI Wei-bo;LIU Xiao-fang(Coikge of Computer Science,Technology,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院

出　　处：《青岛大学学报（自然科学版）》2019年第4期91-95,共5页Journal of Qingdao University(Natural Science Edition)

摘　　要：针对目前智能录播系统中学生站立检测的需求,提出一种基于GoogLeNet神经网络的站立人脸的检测与定位算法。该算法首先利用帧差法和肤色检测初步确定学生站立活动区域,然后在该区域使用迁移学习训练的GoogLeNet神经网络检测是否存在人脸,若存在,则记录人脸的位置信息,最后通过人脸在垂直方向和水平方向上的运动距离来判断是否为站立的学生人脸。实验结果表明,本文方法基本可以实现学生站立人脸的检测与定位功能。Aiming at the demand of students’standing detection in intelligent recording and broadcasting system,this paper implements a detection and localization algorithm based on GoogLeNet neural network.The algorithm firstly uses the frame difference method and skin color detection to initially determine the student’s standing activity area,and then uses the GoogLeNet neural network of the migration learning training to detect whether there is a face in the area.If it exists,the position information of the face is recorded.The distance between the vertical direction and the horizontal direction is used to determine whether it is a standing student’s face.The experimental results show that the method can basically realize the detection and location function of students standing face.

关 键 词：GoogLeNet 卷积神经网络 人脸检测 智能录播系统 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]