Guardian A.N.Y: Detecção algorítmica e classificação de hiperestímulos visuais em vídeos do YouTube para o suporte à acessibilidade sensorial no Transtorno do Espectro Autista(TEA)

Abstract

The consumption of digital media by children has grown significantly, with YouTube standing out as one of the main platforms for access. However, this exposure presents accessibility challenges for children with Autism Spectrum Disorder (ASD), as visual overstimulation, such as rapid cuts and high saturation, can trigger episodes of sensory overload. Although the platform provides features such as captions and screen readers, there remains a gap in the control of sensory stimuli. This study presents the development and technical evaluation of Guardian A.N.Y (Autista No YouTube), a system that operates as an automated filter, categorizing children’s content according to its level of visual stimulation. Based on a dataset manually categorized by the author as a reference for analysis, the system processes videos, extracts frame-level metrics, and applies a Random Forest Classifier algorithm with a stratified cross-validation strategy. The technical tests demonstrated the model’s ability to distinguish between videos with low, moderate, and high levels of visual stimulation, reaching a mean accuracy of 78.46%, suggesting the feasibility of assistive tools focused on the preventive alerting of potentially overstimulating content.