基于MTCNN的人眼状态检测与判断

Human eye state detection and judgment based on MTCNN

针对线上课堂学生听课状态的监督,设计了基于MTCNN的人眼状态检测与判断系统。首先,基于多任务卷积神经网络(MTCNN)算法检测视频中的人脸和人脸特征点,利用人脸和人脸特征点的检测信息定位人眼,然后使用灰度二值化图像处理方法提纯人眼图像,将提纯的人眼图像进行灰度二值化处理,统计虹膜区域像素点的个数和坐标,计算平均坐标,即为虹膜中心点的位置,最后计算提纯人眼中心点与虹膜中心点之间的距离,通过该距离与阈值距离的比较,判断人眼是否关注屏幕两侧。实验结果表明,当学生人脸正面对准屏幕,连续5 s检测到虹膜中心点与人眼中心点之间距离大于阈值距离时,可以判断学生人眼未关注屏幕,即确定学生没有认真听课。

According to the listening state of online classroom students,we design a human eye state detection and judgment system based on MTCNN.Firstly,the face and face feature points in video are detected based on MTCNN;the human eye is located by the detection information of face and face feature points;and then the human eye image is purified by grayscale binary image processing.The purified human eye image is processed by grayscale binarization,counting the number and coordinates of pixels in iris region,calculating the average coordinates that is the position of the iris center point.Finally,the distance between the purified human eye center point and the iris center point is calculated.By comparing the distance with the threshold distance,it is determined whether the human eyes pay attention to both sides of the screen.The experimental results show that when the front face of the student being aligned with the screen and the distance between the iris center point and the human eye center point is greater than the threshold distance for 5 seconds in a row,it can be judged that the student's eyes don't pay attention to the screen,that is to say,the student does not listen carefully.