Automated Exam Invigilator Using Deep Learning

Abstract

From the different physical gestures in the examination room, the examinees' suspicious conduct can be observable. Traditional invigilation methods cannot perform effective monitoring. Due to the physical constraints of human invigilators, many illegal acts go unnoticed. The fundamental principle of this research is to prevent unethical approaches from the examinees by differentiating the correct and suspicious posture of the candidate. An object detection model was developed by YOLOv7(You Only Look Once) to identify the movement of the candidates. At present YOLOv7 has greater accuracy in object detection. During the training, the model has 97.1% mAP(Mean Average Precision). That indicates the high accuracy of the model. After that, we deployed that YOLOv7-trained model into the web with the help of FLASK(Micro Web Framework). With this proposed system, the capability of a large number of students’ invigilation will be increased. Suspicious behavior can be detected in real-time with the help of CCTV footage. The system is completely effective in identifying and keeping track of further than 100 participants in a single frame during assessments. To assess the effectiveness of the Automatic Invigilation System, many authentic examples are taken into account. Any kind of institution can apply this model to examine any kind of candidate. To identify and maintain a close eye on questionable student behavior universities, colleges, and schools might use the suggested invigilation approach. It will reduce academic dishonesty and cheating among students. However, by putting the suggested invigilation mechanism into place, maybe cheating can be stopped and find a remedy for the problem.