Where Are the Children with Autism Looking in Reality?

Xiaoyu Ren,Huiyu Duan,Xiongkuo Min,Yucheng Zhu,Wei Shen,Linlin Wang,Fangyu Shi,Lei Fan,Xiaokang Yang,Guangtao Zhai
DOI: https://doi.org/10.1007/978-3-031-20500-2_48
2022-01-01
Abstract:Social difficulties are hallmarks of individuals with autism spectrum disorder (ASD), of which atypical visual attention is one of the most important characteristics. Learning and modeling the atypical visual attention of individuals with ASD have particularly important significance to related research in the fields of medical science, psychology, education etc., and many studies have been conducted in the literature. However, previous studies have two weaknesses. First, all stimuli in the conducted experiments are selected by the researchers, which are not only restricted by the objective and intention of the researchers, but also limited by the subjective cognition of the photographers. Secondly, most of these stimuli are displayed on screens with restricted and relatively small field-of-view (FOV) compared with the real world. Therefore, in this paper, we conduct the first large-scale study towards better understanding and modeling the atypical visual attention of individuals with ASD in real world. To overcome the two weaknesses mentioned above, a large-scale dataset is established which includes 300 omnidirectional images with the corresponding eye tracking data collected under virtual reality (VR) environment among 15 children with ASD and 16 typically developing (TD) controls. Moreover, a vector quantized saliency prediction model (VQSAL) is applied to better learn the visual attention patterns of both ASD and TD people under the omnidirectional condition.
What problem does this paper attempt to address?