摘要
近年来,人数统计问题在教学场景下的应用需求越来越多,针对目前公开的图像数据集无法满足教室场景下的人脸检测需求,论文提出了包含真实教室场景的图像检测数据集StudentDetection,同时提出了以RetinaFace人脸检测网络为基础进行改进的RetinaStudent人头检测网络,解决了因学生头部姿态导致的脸部被遮挡情况下的人脸识别失败问题,并与当下主流算法进行对比测试,在自制数据集上教室人数统计精确度高达99.1%。
In recent years,there have been more and more applications for people counting problems in teaching scenarios. In view of the fact that currently public image datasets cannot meet the needs of face detection in the classroom scene,this paper establishes StudentDetection dataset for classroom scene image detection and proposes an improved RetinaStudent network based on the RetinaFace network. RetinaStudent migrates the face detection algorithm to the head detection algorithm,which solves the problem of the face being occluded due to the student’s head pose. Camparing RetinaStudent with the current mainstream algorithms,it achieves an accuracy of 99.1% on self-made dataset.
作者
刘媛
陈小丽
屠增辉
谢志敏
郑祎能
LIU Yuan;CHEN Xiaoli;TU Zenghui;XIE Zhimin;ZHENG Yineng(Huazhong University of Science and Technology,Wuhan 430074)
出处
《计算机与数字工程》
2022年第9期1887-1890,1916,共5页
Computer & Digital Engineering
基金
湖北省教学研究项目(编号:2018048)
校级实验技术研究项目(编号:2021-8)资助。
