Classifying interactions of parents and children with Down syndrome in educational environments using deep learning

Recognizing parents’ behaviors is important for the impact on the development of children. This is even more important in parents of children with intellectual disabilities. In this work, we propose using human activity recognition to identify certain behaviors of parents of children with Down syndrome. Specifically, we propose to use computer vision and deep learning to analyze videos of parents and children with Down syndrome interacting in an educational setting for identifying actions related to directive behaviors such as physical interventions. The results obtained through the experiments carried out with the deep learning C3D model show that physical interventions can be recognized with an accuracy greater than 85%. Our results can be used by therapists for identifying actions related to directive behaviors automatically in children with Down syndrome.