Evaluating Gaze Detection for Children with Autism Using the ChildPlay-R Dataset

This study focuses on detecting two-dimensional (2D) gaze points in interactions between children both with and without autism and adults. Within the scope of this objective, a novel subset called ChildPlay-R has been created based on the open-source ChildPlay dataset. This new dataset includes explicitly the interactions of children with and without autism with adults in similar uncontrolled environments and relevant 2D gaze points. Moreover, the analyses using the Modified Spatiotemporal Gaze Detection (M-STGD) model, compared to the traditional Spatiotemporal Gaze Architecture (STGA), reveal significant improvements in the Area Under the Curve values and out-of-sample precision. These advancements indicate the M-STGD model's potential to provide a deeper understanding of the social engagement patterns of children with autism. This study contributes to the automatic detection of 2D gaze points in social interactions and supports the social development of children with autism.