FEATHER: A Low-Restriction Multi-Distance Protocol for Real-World Parkinson’s Disease Assessment

Abstract Accurate and scalable assessment of Parkinson’s disease (PD) remains challenging due to strict data collection requirements and reliance on specialized equipment. We propose FEATHER, a low-restriction and cost-effective multi-distance data collection protocol designed for real-world PD assessment in home and resource-limited settings. Based on this protocol, we curate a open-access FEATHER dataset of 242 finger-tapping videos from 35 clinically diagnosed PD patients, recorded at distances from 30 cm to 500 cm using consumer smartphones. Using a hand detection and pose estimation pipeline, we extract kinematic features for automated score, Movement Disorder Society - Unified Parkinson’s Disease Rating Scale (MDS-UPDRS or UPDRS), prediction and severity classification. Experiments show that recordings at 200 cm achieve the best performance. Our results demonstrate the feasibility of low-barrier, distance-aware PD assessment and support the development of accessible, real-world monitoring systems.