An In-Browser Proctoring System Using YOLO-Based Object Detection and Gaze Analysis

The development of remote learning has increased requirements for effective online evaluations, but the lack of physical supervision can lead to inappropriate actions such as invisible support, illegal collaboration, and the use of restricted resources. This work represents a web-based automated proctoring solution that uses a camera and a lightweight backend system to reduce the need for constant monitoring by avoiding complicated client-side implementation. The proposed system consists of three major components such as a YOLOv8s, a Dlib application and a web-based system. The YOLOv8s vision module detects human faces and restricted objects (such as mobile phones and books), a Dlib facial recognition application estimates eye and head direction to identify behavioral changes and the web-based monitor detects the behaviors like switching tabs, reducing the window size and exiting full-screen mode. A rule-based layer analyzes these signals to determine the type of violation and takes necessary action such as providing real-time notifications and expanding procedures for repeated errors. The detected activities will be saved as structured recorded files in a MySQL database for transparency and scientific evaluation with suitable supporting links shown on administrator dashboards for monitoring after the evaluation. This experimental evaluation and dashboard examples demonstrate that the system will monitor proctoring actions in real-time and provide accurate, reviewable data suitable for remote conditions.