A Novel Autism Spectrum Disorder Children Dataset Based on Thermal Imaging

Autism Spectrum Disorder (ASD) is one of the most common diseases that causes difficulty for a person to express their emotions or to understand the emotions of others. The grouping of autistic children classification of autistic children according to their severity level. Previously, the classification of autistic children has been carried out using visual datasets (RGB images). The RGB images have limitations in relies lighting sources. Therefore, this research provides a dataset of autistic children based on thermal imagery. The advantage of thermal imagery is that it can acquire the facial temperature regardless of lighting conditions and skin color. The proposed dataset consists of 4.080 images with 2.040 images of autistic and normal children. The dataset is taken using a thermal camera FLIR E95. Experimental results show that the proposed dataset result in better classification with an accuracy of 98% for each face detected in real-time. We can conclude that this thermal dataset can be implemented to classify autistic children.