Roadmap¶
Caliscope is a multicamera calibration system for 3D motion capture. It targets researchers seeking low-cost alternatives to commercial systems like Vicon.
Current Priority: Calibration Flexibility¶
The biggest friction point today is the dependency on Charuco boards. For intrinsic calibration, Charuco is overkill — a standard checkerboard suffices and avoids mounting a printed board on a rigid flat surface. For extrinsic calibration, printing a Charuco board large enough to calibrate a big capture volume is impractical.
The plan: checkerboard for intrinsics, ArUco markers for extrinsics.
Phase 1: Checkerboard Intrinsic Calibration¶
Goal: Support standard checkerboards for intrinsic calibration.
Why: Checkerboards are cheap, available in large sizes, and lay flat. Single-camera calibration doesn't require unambiguous rotation determination.
Phase 2: ArUco-based Extrinsic Calibration¶
Goal: Enable multi-camera extrinsic calibration using ArUco markers instead of Charuco boards.
Why: A large ArUco marker on poster paper is visible from distance, cheap to produce, and printable front/back for visibility from any angle. Multiple markers can be scattered around a lab for static calibration. Rigid flatness matters less for extrinsic calibration.
Workflow:
- Scatter ArUco markers around the capture volume
- Navigate to frames showing marker visibility across cameras
- Select one marker as world origin
- Adjust orientation (e.g., marker on bookend defines "up" and floor plane)
- Input known edge lengths to establish scale
This mirrors Vicon workflows with $0.10 printed markers.
Future¶
Tracker Integrations¶
- SLEAP / DeepLabCut: Animal behavior research
- MMPose: Modern human pose estimation
Hardware Integration¶
Frame capture is a separate project. Eventual integration with synchronized camera platforms (webcam clusters, Raspberry Pi arrays) would complete the end-to-end workflow.
Markerless Biomechanics¶
Long-term goal: high-fidelity markerless motion capture for biomechanics research. This requires accuracy sufficient to avoid inferring false forces from position noise, plus reliable joint center estimation for inverse dynamics.