Year: 2024
Type: dissertation
Abstract: Emotion is the internal response of an individual to events, situations, and thoughts they experience. These responses often manifest through physiological changes, such as variations in heart rate, or through body language, such as facial expressions, micro-expressions, and hand-arm gestures. Emotions play a fundamental role in social interaction, as a significant portion of communication between people occurs through emotional expression and reactions. Emotional intelligence, on the other hand, is the ability to understand and differentiate one's own emotions and those of others, and subsequently shape thoughts and behaviors based on this information. Emotional intelligence provides individuals with the capacity for empathy, allowing them to make decisions by considering the feelings and reactions of others. This enhances interpersonal interaction and helps individuals integrate into society. Recently, emotional intelligence has been integrated into humanoid robot-based systems like Pepper, Nao, and Kaspar. These systems are expected to understand the emotions of their counterparts and consider these emotions when deciding on their feedback and behaviors. Research indicates that children with atypical development (such as autism or hearing impairment) may struggle with expressing their emotions and understanding the emotions of others. This can reduce their social adaptation and lead to feelings of isolation. However, interaction with a humanoid robot equipped with an affective computing module could assist these children in socializing and adapting to society. When developing an emotion module for a robotic system, several factors need to be considered. This thesis aims to create an emotion recognition module for social robots using datasets of children with atypical development (autism and hearing impairment). The focus includes determining the types of data to be used for the emotion recognition module, how robots can operate in real-time, and achieving high performance with low cost in the emotion recognition module of robots. Additionally, the questions "Should a single modality be used for the emotion module?" or "Should multiple modalities be used by combining different types of data?" are discussed based on the performance of machine learning models. As a result, this thesis examines three different studies and attempts to answer the previously mentioned questions. In the first study, a real-time emotion recognition module was developed for an assistive social robot system designed for children.
Author: Takır, Şeyma
Institution: Istanbul Technical University
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7117397141
