基于深度学习的教室人数统计系统设计  

Design of Classroom Number Counting System Based on Deep Learning

作　　者：陈久红 张海玉 CHEN Jiu-hong;ZHANG hai-yu(School of Electronics and Communication,Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学电子与信息学院

出　　处：《软件导刊》2019年第10期27-29,35,共4页Software Guide

摘　　要：教室是学生上课和自习的主要场所,由于学校教室有限,寻找无课或人少的教室往往需要花费学生较多时间。开发一个教室人数统计系统,帮助学生快速找到一个合适的自习室很有意义。在教室监控图像中学生个体是小目标,而现有基于卷积神经网络的R-FCN目标检测算法对小目标检测困难。针对这一问题,在R-FCN基础上进行一系列改进,大大提高了R-FCN目标检测算法对小目标的识别能力。在自制的数据集上进行验证,准确率达到了89.4%。Classrooms are the main place for students to attend classes and self-study,due to the limited classrooms in the school,it is often necessary to spend a lot of time for students to look for vacant classrooms for self-study.How to develop a classroom demographic system to help students quickly find a suitable study room is very meaningful.Because the student individual is a small object in the classroom monitoring picture,the existing R-FCN target detection algorithm based on convolutional neural network is difficult to detect small objects,we carried out a series of modifications on the basis of R-FCN,which greatly increases the ability of R-FCN object detection algorithm to identify small targets,thus solving the problem of recognition of students in the classroom.Verified on a self-made data set,the system achieved an accuracy of 89.4%.

关 键 词：人数统计 人头检测 R-FCN 卷积神经网络 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]