Indoor drone mocap without the five-figure invoice

What it does

Low-Cost-Mocap is an outside-in motion capture system: fixed cameras watch a room and compute 3D positions of objects (drones, in practice) inside it. The author uses it to fly multiple quadcopters autonomously indoors — the thing normally reserved for labs with OptiTrack budgets. It’s built from scratch rather than wrapping an existing mocap stack, with a Python backend doing the vision math and a TypeScript web frontend for the interface.

The interesting bit

The pipeline leans on textbook computer vision — epipolar geometry, bundle adjustment, structure-from-motion via OpenCV — rather than learned models. That’s the boring part, and it’s exactly where the value is: classical geometry is cheap, deterministic, and runs on hardware you already own. The author’s related dissertation work goes the opposite direction (inside-out SLAM with cameras on the drones), making this repo the infrastructure-dependent half of a bigger picture.

Key highlights

• Outside-in tracking of multiple agents in a fixed room-scale space
• Uses PS Eye cameras via the pseyepy library — genuinely cheap hardware
• Full stack: Python backend for camera streams and mocap computation, web frontend for visualization
• Camera calibration workflow is community-documented, including a discussion post explaining how camera_params.json is computed
• Companion blog post and video demo cover the drone and camera setup in more depth

Caveats

• The main branch requires OpenCV’s SFM module, which means compiling OpenCV from source — the README itself calls this “a bit of a pain”
• The no-cv-sfm branch removes that dependency but is, by the author’s own admission, completely untested
• Documentation is thin; the README openly asks for help adding type definitions to the Python code

Verdict

Worth a look if you want room-scale tracking for drones or robots without buying a commercial mocap system, and you’re comfortable with OpenCV plumbing. If you want polished docs or a turnkey install, this is a maker project in the honest sense — brilliant, but you’ll be reading source.