Stereotypical motor movements (SMM) are a form of restricted and repetitive behaviors, which are a core symptom of Autism Spectrum Disorder that remain difficult to identify and analyze reliably across individuals and developmental stages. This work presents a novel pipeline that leverages human pose estimation and mathematical tools such as Topological Data Analysis (TDA) to identify and quantify repetitive movement patterns in video-based data. We construct low-dimensional, interpretable feature vectors that capture geometric properties of repetitive motor behavior by extracting time series using video-based pose estimation and analyzing the repetitive movements using TDA. We demonstrate that using these features and simple classifiers enable the accurate classification of SMM and can further generalize to unseen individuals. Using visualization techniques of the feature space reveals that our extracted features generalize across individuals and are not affected by subject-specific mannerisms. Our results highlight the potential of using mathematically principled features to support scalable, interpretable, and person-agnostic detection of stereotypical behaviors in naturalistic settings.

Austin MBaye is a second-year Ph.D. student in the Department of Mathematics at Northeastern University, where he focuses on the theory and applications of Topological Data Analysis (TDA). He recently completed his B.A. in Mathematics at Vassar College in May 2024. Austin’s research centers on the development of mathematically grounded tools for analyzing complex behavioral and structural data, with applications in Autism Spectrum Disorder, optical tomography, and cellular motility. He is actively involved in collaborative, interdisciplinary research at the intersection of mathematics, computational biology, and behavioral science.