Analyzing Human-Robot Interaction Using Machine Vision for Autism screening

Abstract

In this paper, a vision-based approach is proposed to analyze the Human Robot Interaction (HRI) and extract features that can be used to distinguish between children with autism and Typically Developed (TD) children. The algorithm has been tested on the interaction of a set of children with autism and TD children with a parrot-like robot. For each subject, 190 seconds of interaction was recorded, and tracking algorithms were used to determine the position of each child with respect to the robot. Then, features were extracted and Gaussian SVM was used to distinguish between TD and children with autism reaching 81.3% accuracy.