SkellyTracker
SkellyTracker is the single-camera image tracking backend for FreeMoCap. It turns raw images into named 2D keypoints and bounding boxes. It wraps several independent computer-vision tools behind one consistent API built around a Tracker → Session → Detector pipeline, implemented in the skellytracker/core/ package. Part of the FreeMoCap polyrepo's pantheon tier.
It serves two distinct jobs for the pipeline:
- Human pose estimation, MediaPipe (pose/hands/face) or RTMPose + YOLOX produce whole-body keypoints per frame.
- Calibration marker detection, ArUco and ChArUco board detection produce the corner observations FreeMoCap's multi-camera calibration solves against.
Everything operates on single images or batches of images. SkellyTracker knows nothing about cameras, synchronization, recording, or 3D math. SkellyCam supplies frames, and FreeMoCap triangulates the output. Package-wide runtime type checking is enabled via beartype_this_package() in skellytracker/__init__.py.
At a glance
In the FreeMoCap app. Finds the body in every camera frame. Each time FreeMoCap processes a recording, SkellyTracker spots the person, locates their joints (with RTMPose or MediaPipe), and detects the ChArUco board during calibration. Choosing a tracker in the app picks which SkellyTracker detector runs. See choose a tracking model.
| Code | freemocap/skellytracker |
| Issues | Open issues |
| Version in FreeMoCap alpha.25 | 6310005 on main |
| Install | FreeMoCap installs it from GitHub. The PyPI package is an older, separate release; for the current code use uv add git+https://github.com/freemocap/skellytracker |
| License | AGPL-3.0 |