Emotion Recognition of ASD Children using Wavelet Analysis

Autism Spectrum Disorder (ASD) children struggle with social interaction and communication, prompting the development of facial emotion recognition systems to aid communication with caretakers. Researchers have shown interest in using non-invasive thermal imaging to recognize emotions. In this study, a wavelet-based technique was developed to detect changes in the thermal intensity values (TIV) of three regions of interest from frontal facial thermal image. The study analyzed thermal images of ASD children responding to audio-visual stimuli of five basic emotions and derived features from them. Wavelet coefficients were combined with thermal intensity values of each region of interest to form feature set, which were fed into a CNN model. The result showed the efficacy of the method with accuracy and precision attained at 91.81%% and 94.54% respectively. The study is useful in the development of robot assist training for the ASD children as part of early intervention program.