Research On Visual Attention Detection Method Based On Eye Features

The eye is one of the important organs for acquiring external information,and using computer vision methods to detect the direction of human eye gaze can enable machines to better observe and understand human behavioral activities.In this paper,we study the attention detection method based on eye features with the background of online learning,and combine gaze estimation method with the head pose estimation method to achieve the automatic recognition of the attention state of the subject in the face image.For the head pose estimation problem based on face images,an EPnP-based head pose estimation method is used,which first identifies the two-dimensional face key points in the face images and then uses the EPnP method to obtain the angle information of the face in the three-dimensional coordinate system to achieve the head pose estimation.For the face image based gaze estimation problem,a deep learning method is used to improve the existing residual network and propose a gaze estimation method based on a pre-activated residual network,which constructs residual blocks with pre-activation function by setting the BN layer and Re LU before the convolutional layer,and then designs a pre-activated residual network using three pre-activated residual blocks.The network takes the eye image as the input of the model and estimates the human eye sight direction by fusing the head pose information.Finally,the result of fusing the estimation of the subject’s line of sight with the estimation of the head pose is used to implement a vision-based method for attention detection,and a student attention detection system applied to in-school learning is constructed using the above method.In order to verify the effectiveness and robustness of the method in this paper and evaluate the rationality of parameter selection,simulation experiments and performance evaluation are conducted using the publicly available dataset MPIIGaze.The experimental results show that the network model proposed in this paper is more stable than the traditional convolutional neural network and it works better in gaze estimation.Its average angular error is 4.04 degrees,which is smaller and more accurate than other methods,and it has better robustness for low-quality images.In a practical test environment,the method in this paper can achieve automatic recognition of the subject’s head state and attention in images,which can meet the automatic detection of students’ learning state in online learning.