Classroom-Engagement-Detection-using-Facial-Expression-Recognition

Student engagement is a key concept in education. Facial expression of individuals can be utilized as an indicator to detect their engagement level. This project can recognize facial expression from the input video or webcam using Convolutional Neural Network.
FER-2013 dataset from Kaggle is used in training the Deep Learning model. Human accuracy on this FER-2013 dataset is +-65%. Performance evaluation was done using average accuracy on seven classes from the confusion matrix based on the original splitting of dataset.
The highest classification accuracy was 70.44% using ensemble of five deep CNN based on hard voting, followed by 68.99% using a shallow CNN. Some techniques such as haarcascade face detector, data augmentation, mix-max normalization, histogram equalization were used.
A simple web app is built to demonstrate the expression-based engagement detector using Flask as backend and SQLite as database to store the results.