Abstract

The Attentive Proctoring System proposed in this paper addresses the growing need for reliable remote examination solutions amid the global shift toward online learning. Traditional methods of human proctoring are often constrained by scalability issues and resource demands, rendering them inefficient in the face of large-scale online assessments. Leveraging advanced deep learning techniques, our framework aims to ensure exam integrity through a multi-phase approach. By preprocessing video frames captured from students' webcams and employing techniques such as background subtraction and face detection with YOLOv7-SGCN, we establish a robust foundation for identifying potential irregularities. YOLOv7-SGCN is ideally suited for real-time applications since it offers reliable and effective identification of questionable activity with no processing overhead.  However, Attentive-Net improves attention-based feature learning, increasing the precision of recognizing tiny behavioral cues. Further enhancing security measures, our system integrates multi-modal liveness detection and head pose estimation, providing comprehensive monitoring capabilities. Emotion detection, facilitated by a Faster R-CNN, enables the identification of unauthorized aids like mobile devices or books. The integration of Attentive-Net allows for dynamic focus adjustment based on various component outputs, ensuring a thorough examination of pertinent areas within the image. With mechanisms in place for alert and intervention, our system offers a proactive approach to maintaining exam integrity, thereby fostering trust and confidence in the online examination process.

Statistics