摘要
针对传统监考存在人工成本高、主观性强等问题,构建基于人脸识别、头部姿态估计和目标检测的智能监考模型。模型通过人脸识别算法进行考生身份验证,设计结合注意力机制的头部姿态估计(channel and spatial-aware wide head pose estimation network,CS-WHENet)方法对考生偷看的异常行为进行检测,并使用深度学习方法及传统方法对考生传递纸条的异常行为进行联合判定。实验结果表明,智能监考模型在模拟真实考场的环境中,对考生身份验证与异常行为检测均有较高的准确率,并能在GPU支持下实现实时检测。通过验证表明,该模型能有效降低监考人员工作成本,实现考场监考公平性。
Aiming at the problems of high labor cost and strong subjectivity in traditional invigilation,an intelligent invigilation model was constructed based on face recognition,head pose estimation and target detection.The model used face recognition algorithm to authenticate candidates,with a head pose estimation method(channel and spatial-aware wide head pose estimation network,CS-WHENet)combined with the attention mechanism to detect the abnormal behavior of the candidates peeking.Using deep learning methods and traditional methods,the examinee′s pass-through abnormal behavior was jointly determine.In the environment of simulating real examination room,the experimental results indicatd that the intelligent invigilator model could achieve better accuracy of examinee′s identity verification and abnormal behavior detection respectively,and could realize real-time detection with GPU support.Through the verification of the system,the model effectively reduced working cost of invigilators and realized the fairness of invigilators in examination room.
作者
袁欣瑞
王海荣
王振旭
YUAN Xinrui;WANG Hairong;WANG Zhenxu(Department of Computer Science and Engineering,North Minzu University,Yinchuan 750021,China)
出处
《郑州大学学报（理学版）》
CAS
北大核心
2023年第3期41-49,共9页
Journal of Zhengzhou University:Natural Science Edition
基金
宁夏自然科学基金项目(2020AAC03218)
宁夏产教融合人才培养示范专业项目(2018SFZY14)
北方民族大学校级科研项目(2021XYZJK06)
大学生创新项目(2021-XJ-JSJ-010)。
关键词
智能监考
人脸识别
头部姿态估计
目标检测
运动目标检测
intelligent invigilation
face recognition
head pose estimation
target detection
moving target detection
