Markerless video analysis is feasible during a soccer program for autistic children in the SPARK pilot study

Abstract Objective measurement of behavior during real-world interventions in autistic children remains challenging. This prospective, single-arm pilot study assessed the feasibility of markerless video analysis during an eight-week structured soccer program. Ten boys aged 7.5–10 years with level 1 autism completed eight weekly 60-minute sessions. Recordings from sessions 1 and 8 were processed using two-dimensional pose estimation to extract seven candidate motor and socio-motor metrics. Paired recordings were analyzable in all participants. HeadUp ratio, Posture Stability, Movement Fluidity, Active Time, Proximity, and Orientation increased in every child, whereas Restlessness decreased (exact two-sided Wilcoxon p = 0.00195 for each metric; all significant after false-discovery-rate correction). Among paired caregiver questionnaires, mean social responsiveness raw scores decreased from 97.22 to 76.78 and parenting stress scores from 100.11 to 85.56 (p = 0.0117 for each). Digital and questionnaire changes were directionally congruent, without formal association testing. Markerless analysis provided feasible, low-burden longitudinal measurement during naturalistic activities in autistic children. The uncontrolled design and absence of formal measurement validation preclude conclusions about intervention efficacy, reliability, or construct validity. Further validation is required before these metrics can serve as established clinical outcomes.