Computer vision mobility assessment
An iPad app measuring shoulder range of motion from camera pose data, for a clinical research pilot.
The repository arrived stalled: the working code sat on a branch six weeks ahead of main, behind four pose engines that did not run and 66 MB of licence-encumbered binaries. I established a contractual baseline, tore the dead stack out, and rebuilt guided capture for flexion, abduction and hand-to-head across both sides. Reporting on screen and as PDF, research data export, and a written operator guide. I moved the design local-first, so participant data never leaves the lab iPad — a stronger privacy position than the hosted approach originally scoped, and one less thing for the ethics board to weigh.