Deploying A Standalone Facial and Emotion Recognition Classroom Management System on Resource-Constrained.

Abstract

In recent times, it has been proven in most industries that deep learning can play a huge part in the development and automation of processes otherwise performed manually by humans alone. The trend however has encountered more of a shift and tend towards transfer learning where standalone systems can be built on weights that have been extensively trained to be use-case agnostic. This project seeks to address the problem of student truancy. The methodology applied is a combination of a deep learning use-case agnostic weight embedding obtained from a popular network called Face net. Recognition is performed by computing facial distances using the weight embedding. Also addressed is the common reliance on internet for functionality present in most modern-day systems by deploying all the resources necessary on a resource-constrained development board. Emotions during class are also analyzed to improve classroom experience which will be displayed on a web application dashboard powered by artificial intelligence back-end. The results obtained show an above average recognition rate of 0.63 with emotional recognition accuracy of 0.72. The implications of these results are that accurate attendance can be taken in an organization with minor increments to the system such as increased computational capabilities.