An improved facial emotion recognition system using convolutional neural network for the optimization of human robot...

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Machine Learning

Background:

• Robotics applications increasingly require sophisticated human-robot interactions (HRI) for effective functionality.
• Computer vision, particularly facial emotion recognition (FER), is crucial for enhancing HRI precision.
• Identifying human emotional states is key to developing more intuitive and responsive robotic systems.

Purpose of the Study:

• To develop and evaluate software capable of recognizing human emotions from images and videos using computer vision and machine learning (ML).
• To investigate the feasibility of training software based on emotional expressions for FER.
• To identify essential facial gestures for CNN-based FER systems.

Main Methods:

• Utilized ML techniques and a digital image processing pipeline for emotion recognition.
• Developed and trained software incorporating convolutional neural networks (CNNs) for FER.
• Compared the proposed model's performance against established datasets: FER2013, RAF-DB, and CK+.

Main Results:

• The developed software demonstrated the feasibility of recognizing emotions in human gestures.
• Achieved a high accuracy rate of approximately 95% on the CK+ dataset.
• Performance varied across datasets, with the FER2013 dataset yielding the lowest accuracy at around 64%.

Conclusions:

• The study confirms the viability of CNNs for accurate facial emotion recognition.
• The findings contribute to advancing knowledge in neural networks and improving computer vision efficiency.
• Enhanced FER capabilities can significantly improve the quality of human-robot interactions in various applications.