Autistic Verbal Behavior Parameters

Abstract

When patients suffer ASD (Autism Spectrum Disorder), they experience severe difficulties in speech and socializing. Even a simple sentence and facial expressions are hard to understand for them. Biotech is a research project to get new alternates to help these individuals to communicate. The main goal here is to provide advances and tuned tools through audio and video real-time processing. Part of the previous work in this research stated the basis to understand that every individual has its own natural language expressed in ways of Robert Berwicks’ verbal communications. Rather than only consider a one way approach on traditional PECS, the contribution of the entire research is to articulate some kind of feedback based on the way the patient reacts upon each event. This makes it possible to get the inner feeling of the individual. This paper is focused on the performance and evaluation of head bouncing and can be extended to some other movements during interactions recorded in diverse circumstances with ASD patients. A short tracking and lightweight processing approach is presented, and applied to a small video test set. Results are validated against manually detected movements. A workflow for processing stimmings and collecting relevant meta data is presented. It applies Machine Learning, and represents a model for tuning the tracking process. Preliminary results show that precision is good enough for detecting the main patient’s behavior and for collecting information about the set of specific individual’s autistic verbal behavior.