Abordagem computacional para aprimoramento das habilidades com as emoções em indivíduos com autismo

Abstract

The detection of facial expressions and the recognition of basic emotions are important in the interpersonal relationships of individuals in a society. Neurotypical individuals have these skills enhanced naturally and gradually during the life cycle. However, individuals with Autism Spectrum Disorder (ASD) have difficulties with the detection of facial expressions and representation of emotions. This restriction affects the social interactions of individuals with ASD in a social environment. This work presents a computational tool that employs serious games to aid in the learning and improvement of these skills in individuals with ASD. The computational algorithms for emotion detection and recognition were developed in an approach that combines manual and learned feature descriptors, by a convolutional neural network model. This tool uses concepts from serious games, 3D modeling for characters, scenarios, and game-based learning methodologies, in order to provide the development of these skills in a playful way. This tool works on several platforms, running online, for interaction with the user. For the detection and recognition phase, four public domain image databases (CK+, FER2013, RAF-DB, and MMI) were used for evaluation. The game was investigated with volunteers in five sessions for the research phases (reference, intervention, and maintenance). The results showed that the proposed tool contributed to the improvement of basic emotion detection and recognition skills in individuals with ASD.