基于教室监控视频的学生位置检测和人脸图像捕获算法

Student Position Detection and Face Image Capture Algorithm Based on Classroom Monitoring Video

实现一个基于课堂监控视频的学生位置检测和学生人脸图像获取系统。本系统由一个定焦全景摄像机和一个PTZ(平移(Pan)、倾斜(Tilt)、变焦(Zoom))摄像机组成。首先利用全景摄像机获得教室全景图像,针对实际课堂环境中的光线突变,提出基于帧间差的异常光线排除算法,实现异常光线监测和动态空教室图像检测与存储;使用HR网络结构对全景图像进行人脸检测,得到人脸检测框集合;针对非约束环境中学生因姿势变化和人脸遮挡、全景图像分辨率低等因素引起人脸信息缺失而导致人脸检测漏检问题,提出基于人体头肩特征的加权运动目标检测算法,得到目标检测框,提高人脸信息缺失的学生位置检测率;针对多种检测框的大量冗余,提出多种检测框加权融合算法,有效减少检测框的重复,得到学生人物检测框集合。然后,将学生人物检测框包含的位置信息传递至PTZ摄像机控制子系统,使PTZ逐个聚焦目标学生,捕获到清晰的学生人脸图像,为后续的人脸识别提供高质量的图像。

This paper implements a student position detection and face image acquisition system based on classroom monitoring video.The system consists of a fixed focus panoramic camera and a PTZ camera.At first,the fixed camera obtains a panoramic image of the actual classroom environment.An abnormal light elimination algorithm based on the frame difference is proposed to realize the abnormal light and empty classroom detection and background image storage.The face detection is realized by HR network structure,which obtains a set of face detection boxes.As the lack of face information caused by the change of posture,occlusion and low resolution,this paper proposes a weighted moving target detection algorithm based on human head and shoulder feature to get target detection boxes,improving the position detection rate of students without face information.Aiming at the large amount of redundancy of multiple detection frames,this paper proposes multiple detection frame weighted fusion algorithms to effectively reduce the detection boxes.After the repetition of the test frame,the set of student character test frame is obtained.Then,the position information contained by the position detection is transmitted to the PTZ camera control subsystem,so that the PTZ focuses on the target students one by one,capturing a clear student face image,and providing a high quality image for face recognition.