Processing and Analysis of Eye Movement Signals for Autistic Children in Omnidirectional Space

‍ ‍To address the problem that previous eye-movement studies for autism spectrum disorder （ASD） cannot show the visual attention of autistic patients in the real omnidirectional world， in this paper， we propose to utilize virtual reality （VR） technology and omnidirectional 360 degree images to collect eye movement data and head movement data in VR environment， and then further analyze and simulate the visual attention of autistic children in the real omnidirectional world. Specifically， we first establish a large-scale panoramic image eye-movement dataset for autistic children. Then， based on the constructed dataset， we improve the extracted features and methods in the three-layer saliency calculation model. We quantify the features of all panoramic images of the dataset， and calculate the influence of different features on visual attention via support vector machine （SVM）. Finally， based on the collected data and computational results， we qualitatively and quantitatively analyze the correlation and difference between the visual attention of autistic children and typically developing controls including the visual preference for different features and head-eye movement correlations. This study can contribute to analyzing the visual characteristics of autism and can further assist the procedure of classification， diagnosis and rehabilitation of autism.