Hello everyone,
Happy to share the Demo of RehabVision :
The Problem
Physiotherapy recovery is tough — not because the exercises are complicated, but because patients are often left on their own between clinic visits. Did you do the exercise correctly? Enough reps? With proper form? Most people just guess. Meanwhile, existing solutions either require cloud processing (raising privacy concerns with sensitive health data) or expensive specialist equipment.
The Idea
RehabVision is a local, vision-based physiotherapy assistant powered by the Metis chip from Axelera AI. It uses a camera to observe a patient performing their prescribed exercises, detects body poses and movements in real-time, and provides feedback on form, repetition counting, and progress tracking — all running entirely on-device.
No cloud. No data leaving your space. Just a smart assistant watching your recovery and helping you get better, right in your living room.
Why Edge AI Matters Here
Health data is some of the most personal data there is. Patients recovering from injuries or surgeries shouldn't have to choose between getting AI-powered guidance and giving up their privacy. By running everything locally on Metis, RehabVision keeps video and pose data exactly where it belongs — on the device, in the patient's own space.
The low-latency inference on Metis also means real-time feedback is actually real-time. No waiting for a server round-trip while you're mid-squat wondering if your knee alignment is right.
How It Works
- Camera captures the patient performing exercises
- Pose estimation model runs on Metis to detect body keypoints
- Exercise patterns are matched against predefined movement templates
- Real-time feedback is provided on form correctness and rep counting
- All processing stays local — fully offline capable
Code - https://github.com/dannytata8041-dotcom/rehabVision/tree/main
I'd love to expand this with more exercise types, a simple UI for physiotherapists to define custom exercise programs, and potentially integration with wearable sensors for even richer feedback. The Metis platform makes this kind of expansion very feasible given its performance headroom.
Thanks to the Axelera team for putting together this challenge — it's been a great experience building something that combines edge AI with real-world health impact. Looking forward to seeing everyone else's projects!