Development of an examination hall Malpractice detection system using pose Estimation and machine learning

Examination malpractice remains a critical concern, driving the need for robust and efficient detection systems to uphold academic integrity. This research introduces a novel approach to examination malpractice detection by leveraging pose estimation and machine learning to address key gaps in existing solutions. Current systems often rely exclusively on either pose estimation or object detection, failing to exploit the synergistic potential of combining these techniques. This study bridges this gap by integrating deep learning models with complementary capabilities to enhance detection accuracy, robustness, and coverage. The proposed methodology utilises OpenPose for precise human pose estimation, enabling the detection of body movements, gestures, and spatial dynamics, alongside YOLO for efficient real-time object detection. This integration allows the system to not only monitor human poses but also detect and analyse relevant objects or cheating aids, such as unauthorised devices or written materials. The fusion of these techniques addresses the limitations of standalone approaches, enabling the detection of cheating scenarios in dynamic examination environments. The system was rigorously evaluated using a curated dataset of diverse examination settings, achieving impressive performance metrics: accuracy of 91.71%, precision of 90.74%, F1-score of 94.07%, and recall of 97.64%. Furthermore, the developed system was seamlessly integrated into a web application enabling real-time use by institutions and users. The integration of advanced computer vision techniques, machine learning algorithms, and modern web development technologies has resulted in a user-friendly, intuitive, and efficient tool for maintaining the integrity of examinations. This system demonstrates significant improvements in detection performance over existing solutions.