The exponential growth of behavioral disorders among human beings is an increasing concern in the medical community due to a lack of medical resources for early identification of atypical behaviors. Many psychological journals which address behavioral disorders indicate that the most prominent way of puzzling out this problem is to identify behavioral disorder characteristics such as repetitive behaviors in early childhood, and recover through therapy. Even though there are many standards presently used in diagnosing behavioral disorder characteristics, due to a lack of facilities and limited number of professionals in the relevant fields, these traditional standards have failed to cater the increasing number of cases with behavioral disorders. Hence, the need for developing automated approaches to overcome the problems in current systems of diagnosing children with behavioral disorders has arisen from the research community. Therefore, the purpose of this study is to develop an automated model that analyzes a video to distinguish typical, and atypical repetitive head movements of children while using different learning methods to mitigate issues that affect the performance of the model due to the scarcity of child datasets. In this work, we present a fusion of transformer networks, and Non-deterministic Finite Automata (NFA) techniques, which classify repetitive head movements of a child as typical or atypical based on the analysis of gender, age, the type of the repetitive head movement along with count, duration, and frequency of each repetitive head movement. Different transfer learning methods were experimented to enhance the performance of the model. The experimental results on five datasets: NIR face dataset, Bosphorus 3D face dataset, ASD Dataset, SSBD dataset, and head movements in the wild dataset, corroborate our proposed model has outperformed many state-of-the-art frameworks when distinguishing typical and atypical repetitive head movements of children.

No abstract

No abstract