ASIC-Based Facial Emotion Recognition System for Human-Computer Interaction

Abstract

In the era of technology, Artificial Intelligence (AI) plays a vital role in human daily life, especially AI machines. But there are some limitations when AI machines communicate with the humans such as recognition accuracy and recognition speed. In this study, the Facial Emotion Recognition (FER) System that used Convolution Neural Network (CNN) technique is developed using Application Specific Integrated Circuit (ASIC) implementation. The free online software – Google Colab is used to train the CNN deep learning model and generate the weight and bias values that are used during the designing of the CNN model in Verilog Hardware Description Language (HDL). The facial emotion expression images undergo the trained CNN deep learning model to classify into seven basic universal emotions such as happy, sad, angry, disgust, surprise, fear and neutral. Functional verification, logic synthesis and physical synthesis are carried out using the Electronic Design Automation (EDA) tool - Synopsys with 32nm technology. The final layout of the FER system had an area of 3101.20 〖μm〗^2 with a total power consumption of 871.05 μW. This system is achieving 92% of recognition accuracy and used 133767ns to recognize an image and classify it according to the emotion class. The proposed design with a compact size and low power consumption can be beneficial for various automation applications such as human-computer interaction systems.