An Integrated AI System for Exam Proctoring Using YOLOv8, Face Recognition, and Noise Detection

Abstract

Online examinations have become increasingly prevalent, demanding robust proctoring solutions to ensure integrity and prevent unauthorized activities. This paper presents an integrated AI-based proctoring system that combines real-time face recognition,
object detection using YOLOv8, and audio anomaly detection to enhance the reliability and fairness of online assessments. The proposed system authenticates candidates through webcam-based face recognition using facial encodings, preventing impersonation. Once verified, the system continuously monitors the candidate throughout the exam session. A YOLOv8 deep learning model is employed to detect the presence of unauthorized objects such asmobile phones, laptops, or multiple persons in the webcam frame.  Additionally, Haar cascades are used to track head and eye direction, flagging suspicious behavior such as looking away from the screen. Simultaneously, an audio detection module monitors environmental noise levels to detect conversations or abnormal sounds, triggering real-time alerts. Detected anomalies activate a visual warning system, sound an alarm, and log the event through automatic
screenshot capture and video recording. All sessions and alerts are stored for later review. This multimodal approach enhances exam security and provides scalable, automated proctoring suitable for remote education environments.