SafeSpace Coach
Computer Vision coaching for athletes
Overview
As a competitive volleyball player, I know how hard it is to get quality feedback on your form without a coach watching. SafeSpace Coach uses MediaPipe's pose estimation to analyse exercises in real time and give instant, private feedback — no gym, no judgment.
Stack
Preview
Add screenshot 1
Drop an image here
Add screenshot 2
Drop an image here
What I learned
- Understood MediaPipe's pose estimation model and how to map 33 skeleton keypoints to angles
- Built real-time angle calculation for squat depth, plank alignment, and lunge form
- Learned the gap between a working demo and a usable product — latency matters enormously
- Applied sports science domain knowledge directly to technical design decisions
Build log
Struggles, findings, decisions, breakthroughs — the honest story.
33 keypoints, which ones matter?
MediaPipe gives you the full skeleton. Figuring out the minimal set of keypoints for accurate squat depth calculation took a lot of trial and iteration with a tape measure.
Lighting kills accuracy
Poor lighting tanks pose estimation confidence. Had to add a low-confidence warning so users know when the camera can't reliably track them.
Vector angles over pixel distances
Early attempts used pixel distances to measure form. Switching to vector angle calculations made it camera-distance-independent and much more reliable.
No cloud, no data
Processing everything locally was a deliberate choice. Athlete body footage is sensitive. SafeSpace = safe literally — nothing leaves the device.