Clinical question
Can face redaction become fast and repeatable without losing the ability to extract movement features?
Study design
SecurePose was evaluated using outpatient gait videos of 116 children with cerebral palsy. The researchers compared automated face detection and redaction against robust manual processing and evaluated intermediate kinematics extraction.
What the researchers found
The system outperformed six tested automated face-detection approaches, achieved redaction comparable with robust manual blurring and completed the process 91.08% faster than the manual method.
Why this matters for MedMeters
This is especially relevant to MedMeters’ planned privacy-first capture flow, where minimizing identifiable video is a design requirement before downstream measurement and sharing.
What this does not prove
Face blurring does not amount to complete HIPAA de-identification. Body movement, voice, room features and metadata can still identify a person. The study does not establish that SecurePose ran entirely on-device, or that MedMeters already has these capabilities.
SecurePose: Automated face blurring and human movement kinematics extraction from videos recorded in clinical settings
Bajpai & Aravamuthan · Scientific Reports · 2026
DOI: 10.1038/s41598-026-50340-5