Desarrollo de un Algoritmo de Visión Artificial para detectar Plagio en Universidades

In this study, we present the development of a computer vision algorithm to detect suspicious behaviors associated with academic plagiarism during in-person assessments at universities. Using convolutional neural networks (CVNs) and tools such as Python, MediaPipe, OpenCV, and YOLOv8, the system was designed to identify actions such as cell phone use, prolonged sideways glances, or the use of real-time notes. The algorithm was trained with a database of 7,136 simulated and labeled images, achieving an accuracy of 87%, a recall rate of 83.9%, and a false alarm rate of 8%. Additionally, metrics such as F1-score, SSIM, and IoU were applied to validate the system's robustness to variations in lighting and motion. The implementation of this system strengthens academic monitoring, promotes fairness in assessments, and offers a replicable and scalable solution for other educational institutions. Ultimately, the combination of advanced technologies allows for improved monitoring mechanisms without replacing human judgment, contributing to educational ethics and transparency.