Recognizing Challenging Behavior for Multiple Children with Intellectual and Developmental Disabilities Using the 3SLC Method

Challenging behaviors in children with IDD(Intellectual and Developmental Disabilities) can threaten their safety and the safety of others as well as the surrounding environment. Recording challenging behaviors in IDD children is currently done manually, which may lead to errors. We generate videos for each person in a multi-user setting and extract joint coordinate values to create composite joint information. In this paper, we propose the 3SLC method that combines Self - Attention, Long-Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN). The proposed 3SLC method analyzes the relationships between the joint composite information, extracts temporal features from the feature map, and analyzes spatial characteristics to predict the behaviors of children with IDD.