Automated proctoring systems are increasingly being adopted in remote examination environments to detect academic misconduct. However, most existing software-based solutions have a high false positive rate and limited verification mechanisms, reducing overall reliability. This paper proposes a lightweight multi-stage automated exam monitoring framework that integrates real-time posture detection with a secondary head yaw-angle verification to improve cheating detection accuracy without requiring additional hardware. The proposed system employs FaceNet-based facial recognition for student identification, ensuring secure enrollment and identity verification. For behavioral monitoring, YOLOv8n-pose estimation is utilized to detect body posture and key anatomical landmarks in real-time. In order to reduce false alarms, suspicious posture events detected in the first stage are further validated using a Perspective-n-Point (PnP)-based head pose estimation method to compute yaw angles and apply threshold-based verification, enabling more robust classification. The experimental results show that the proposed framework achieves 98.33% accuracy with strong precision, recall, and F1-score, demonstrating reliable real-time examination monitoring. By combining facial recognition, posture estimation, and yaw-angle verification in a multi-stage pipeline, false positives are significantly reduced while maintaining computational efficiency, making the system suitable for practical classroom deployment.