Enhancing Examination Integrity Through Vision-Based Smart Proctoring Powered by YOLOv8 and Dlib frontal face detector.

The ubiquity of online exam systems, driven by the swift surge in digitization and globalization, especially in times of global health crises like pandemics, necessitates a robust cheating detection mechanism. While current platforms employ conventional machine learning techniques based on vision, their efficacy is limited by the reliance on manually crafted features and the inability to extract hierarchical representations from raw data. This paper aims to address challenges related to complex system integration and low-resolution video by introducing a practical and efficient deep learning-based method for real-time cheating detection in online exam systems. Leveraging recorded videos from a custom real-world video dataset, YOLOv8 is used for object detection, while Dlib serves as a frontal face detector using a Histogram of Oriented Gradients (HOG)-based approach. The proposed method fuses the outputs of YOLOv8 and HOG to accurately detect faces, enhancing detection reliability and performance. The proposed method, which combines YOLOv8 and Dlib’s HOG-based detector, achieved exceptionally high performance in the fused system, reaching 99.9% accuracy, 99.8% precision, 99.7% recall, and a 99.6% F1-score, surpassing existing methods. This fused approach accurately detected faces in online exam hall scenarios, demonstrating its reliability and suitability for maintaining exam integrity.