Real-Time Facial Emotion Detection Through the Use of Machine Learning and On-Edge Computing

In this research study, we implemented a customized deep learning model for sentiment analysis through facial emotion detection to be used in real-time. This study aims to maximize the model's accuracy and create a lightweight model for real-time Facial Expression Recognition (FER) on edge devices. In order to accomplish this goal, we researched the most recent models and techniques used for FER. We developed our fine-tuned model using FER2013, AffectNet, JAFFE, CK+, and KDEF datasets with 87.0% accuracy, comparable to the most accurate real-time models today, which range from 65-75% accuracy. The primary advantage of the proposed model is the simplified architecture which makes it lightweight and suitable to be deployed on various edge devices for real-time applications. We show how a lightweight but fine-tuned model can achieve higher accuracy than more complicated models proposed in recent studies.