ARAIS-Activity Recognition of Autism Children During Intervention Sessions

Autism Spectrum Disorder (ASD) is a neurodevelopment disorder, which may bring various social and behavioral challenges in a child's life. ASD affects children's overall growth and development by limiting their social and cognitive skills. Play therapy (PT) for ASD is the most common type of intervention that is given by trained therapists to improve the social, communicational, and behavioral skills in children and is mostly analyzed manually. To achieve the positive results of the intervention, it is necessary to monitor and analyze all the activities performed by the children during therapy sessions. In this paper, we have proposed two frameworks for activity recognition of autistic children during intervention sessions, based on a Long Short-Term Memory (LSTM) network i.e. 2D-ARAISLSTM and 3D-ARAIS-LSTM. In these frameworks, 2D and 3D skeleton movements have been used as input modalities for recognizing the complex tasks/activities performed during intervention sessions. The recognition accuracy of the proposed 2D-ARAIS-LSTM model is 95.72% for the 2D skeleton data. Whereas the accuracy of the proposed 3D-ARAIS-LSTM model is 85.00% for the 3D skeleton, on the multimodal-autism intervention dataset (MMASD).