Measuring the Variabilities in the Body Postures of the Children for Early Detection of Autism Spectrum Disorder (ASD)

摘要

Presently, the number of children with autism appears to be growing at disturbing rate. Unfortunately, the awareness of early sign of Autism Spectrum Disorder (ASD) is still insufficiently provided to the public. Arm flapping is a good example of a stereotypical behavior of ASD early sign. Typically, a standard Repetitive Behavior Scale-Revised (RBSR) - set of questionnaire -used by clinicians for ASD diagnosis usually involved multiple and long sessions that apparendy would delay and may have nonconformity. Thus, we aim to propose a computational framework to semi-automate the diagnosis process. We used human action recognition (HAR) algorithm. HAR involved in human body detection and the skeleton representation to show the arm asymmetrical in arm flapping movement which indicates the possibility of ASD signs by extracting the body pose into stickman model. The proposed framework has been tested against the video clips of children performing arm flapping behavior taken from public dataset. The outcome of this study is expected to detect early sign of ASD based on asymmetry measurement of arm flapping behavior.