Assessment of Independence in Passing Entrance Online Testing by Analyzing the Applicant's Head Rotation Angle Based on Neural Network Image Recognition Methods

The work is devoted to the construction of a model for assessing the independence of applicants in passing an online entrance test based on an analysis of the direction of their gaze using neural network image recognition methods. The normalized deviation in the rating of this person before and after enrollment was taken as a metric for assessing the applicant's independence in passing the online entrance test in mathematics. All experiments presented in the work were carried out on real data. Our collection of images was an array of frames from web cameras of553 applicants while they were taking an online entrance exam in mathematics. As part of data preprocessing, images with poor quality were removed from the collection, as well as images where there were two or more people in the frame, or the applicant's head did not fit completely into the frame (less than 90 %). The final size of the collection was 3828 examples. The RealHePoNet convolutional neural network (CNN), proposed by Spanish researchers, was used as a detector of the incoming person's head and a method for determining the direction of his gaze. The accuracy of detecting the head of incoming images from our collection was 82.89 %. By analyzing the area and aspect ratio of the rectangles that, according to the RealHePoNet CNN, limit the incoming head, we managed to increase the detection accuracy to 97.81 %. When the average value of the incoming head rotation angle exceeds the threshold of 6.8°, the average normalized deviation in the rating before and after enrollment becomes negative, and when it exceeds 10.2°, this deviation is less than —2 points, which indicates an increased likelihood of being recruited to take the exam strangers. The average angle of the applicant's head tilt is over 46°, just like the head lift over 18°, may also indicate fraudulent actions on the part of the person taking the exam.