Estimating a climber's Center of Mass — fusing a sensorized wall and monocular video
Politecnico di Milano · Project ACCEPT (Adaptive Climbing for Cerebral Palsy)
Built a marker-less, out-of-lab method to track a climber's 3D Center of Mass during real climbs — replacing expensive, intrusive Vicon-style motion-capture rigs with a single camera + a sensorized climbing wall.
The system fuses two data streams: a neural network (Mediapipe) extracts the climber's 2D pose from monocular video, while tri-axial load cells embedded in the holds capture the 3D force vector at each grip. A weighted geometric model (Zatsiorsky anthropometric tables) reconstructs the body Center of Mass; force data fills in the depth coordinate the camera can't see.
Validated against a Vicon optoelectronic system (gold standard) across standing, overhead-squat and lateral-squat tasks. Final accuracy: ~1.5 cm on x/y, ~5 cm on z — competitive with lab-grade systems, but usable in real gyms with kids in rehabilitation.
Originally designed to help children with cerebral palsy benefit from climbing-based rehab — the same tooling generalises to any sport-tech application that needs marker-less performance tracking.
