Development of computer vision based assistive software for accurate analysis of autistic child stereotypic behavior

Many children with autism most frequently exhibit stereotypical movements (SM) such as arm flapping, spinning, body rocking, etc., which are controllable at early stages with proper detection and treatment procedure. Though, the detection of the SM relies on careful observations and time consuming manual quantification, an intelligent assistive diagnostic tool to capture the SM could aid the therapist for better provision of treatment. This study presents a stereotypic movement detection and analyzing (SMDA) software that apprehended SM of selected body part or whole body in terms of intensity and frequency. The SMDA is powered by (i) computer vision (CV), a video pre-processing module; and (ii) media pipe library (MPL) with a machine learning (ML) framework, to process recorded videos and provide vital information to therapist. Using CV and MPL, landmarks of body parts (LBP) are identified and subsequently labeled them. Further, the labelled landmarks (example: wrist) are tracked during the onset of SM in recorded videos and extracted the X- Y coordinates. Using, data peak filtering algorithm (DPFA), the intensity of SM and its frequency are estimated. Quick detection of vital parameters of SM and versatile LBP selection, features SMDA standout to be a better diagnostic assistive tool to the therapist.