Exam Fraud Prevention and Class Concentration Improvement System for an Online Environment
Min Soo Ko, Heeyoung Joo, Hyok Jae Song · 2022
This paper presents exam fraud prevention and class concentration improvement system for an online environment. The proposed system consists of deep learning-based face analysis, video analysis, and action recognition modules. The face analysis module first detects a face and performs face recognition for identification. Then, face spoofing detection is performed to determine if it is a fraud face, and gaze estimation and facial expression recognition are conducted to inspect the student's condition. And the video analysis module identifies whether the current image is the correct online class or exam state and detects inappropriate objects around it. Finally, the action recognition module determines abnormal online class or exam behavior. For performing on client devices., all deep learning models are optimized and converted through ONNX to be operated in an internet browser environment. The proposed system required about 300ms to analyze an image in a typical laptop computer.