Facial verification in assessments using deep convolutional neural networks
Online assessments are increasingly present. This is even more evident with the great demand resulting from the 2020 pandemic. However, verifying the authenticity of the evaluation is a task that still lacks new solutions. In this sense, this work presents a plugin for Moodle, a Learning Management System widely used at different levels of education. Trained deep convolutional neural networks were used to quickly detect the contour of the face using pixels from the eyes, nose, and mouth. Another network was used for facial verification, extracting a 128-dimension vector from each face using the ResNet-34 neural network architecture. The plugin was tested with 32 users. The results show the feasibility of using this plugin with an average approval rate of 76% through a questionnaire applied to 16 users in optional online activities carried out in Moodle in late 2020 and early 2021. Despite the challenges presented by the pandemic, the solutions presented can be useful after this period of isolation.