Automated Proctoring Based on Head Orientation Analysis and Object Detection

Evaluation of a student's understanding, learning, and ability to apply the learned knowledge for solving real-life problems is one of the critical components in any educational program. A continuous possibility of candidate's involvement in cheating using adaptive digital technologies throughout the assessment of the course curriculum, including examinations makes the detection and prevention of cheating difficult. Due to the prolonged COVID-19 pandemic and smooth continuation of educational programs, key examinations for evaluating students progress need to be done online. Additionally, with the availability of low-cost Internet facility and ever-growing online study materials, online examinations held with pen and paper without constant invigilation are susceptible to cheating. In this paper, we propose a low-cost online proctoring model that identifies cheating during online examinations. Initially, we have identified key cheating behaviors. Based on that, we have build a model which performs live video-based automated proctoring by video analysis and cheating detection using a web camera. The model has three components, student's vision estimation for monitoring head orientation, head count estimation for calculating people in the room, and object/phone detection for calculating phone confidence level. In next phase, these three parameters are combined and sent as input into a trained model that reaches a confidence level suggestive of the cheating behavior of the student. In this work, we have used SVM, decision tree and linear regression models for computing confidence levels. The performance of each model has been discussed and compared in the paper.