Caliscope

Multicamera Calibration + Pose Estimation --> Open Source Motion Capture

About

Caliscope is a GUI-based multicamera calibration package. When the intrinsic (focal length, optical center, and distortion) as well as extrinsic (relative rotation and translation) properties of a set of cameras are known, synchronized frames from those cameras can be used to triangulate landmarks identified across their multiple points of view. With more cameras, this 3D tracking becomes more robust to occlusion and the inevitable errors in 2D landmark tracking and camera property estimates.

While OpenCV provides straightforward functions for the estimation of single camera intrinsic properties as well as estimates of the extrinsic properties of two cameras, there is no straightforward way to estimate extrinsic properties for more than two cameras. Performing this requires bundle adjustment, which demands an extensive series of computational steps and intermediate data tracking.

Caliscope automates this more complex calibration function along with providing visual feedback regarding parameter estimates at each stage of processing. Additionally, there are sample implementations of a Tracker class using Google's Mediapipe that demonstrate the capacity to integrate the full calibration results with landmark tracking tools to achieve 3D pose estimation. While Mediapipe pose estimation has limitations regarding accuracy and precision, it demonstrates a data processing pipeline that can easily integrate more powerful tracking tools as they emerge.

This README provides a general overview and quick guide to install Caliscope on your system. For a more detailed description of the calibration process and workflow, please see our docs.

For a quick sense of how the calibration process works, check out the video walk through of the sample project.

---

Demo Animation

ballet_demo.mp4

Caliscope was used to calibrate the cameras used in the above recordings (both intrinsic and extrinsic parameters), run 2D landmark estimation across all the videos using Holistic Mediapipe pose estimation, then triangulate the 3D landmark positions based on the camera system calibration. The animation was created in Blender based on these 3D landmark estimates using a companion project called Rigmarole. The concurrent video was recorded with a 7 year old t440p laptop, 4 webcams (~$25 each) and another project called MultiWebCam.

---

Quick Start

Basic installation instructions can be found here. Please note that given the size of some core dependencies (OpenCV, Mediapipe, and PySide6 are among them) installation and initial launch can take a while.

For a basic overview of the entire workflow, please see the sample project for an example dataset and a video demonstration of the complete workflow.

Key Features

Calibration board creation

• Easy creation of png files for ChArUco calibration boards
• board definition can be changed across intrinsic and extrinsic calibration allowing greater flexibiltiy

Intrinsic Camera Calibration

• Automated calculation of camera intrinsic properties from input video

  • Optical Center
  • Focal Length
  • Lens Distortion

• Visualization of distortion model to ensure reasonableness

Extrinsic Camera Calibration

• Automated bundle adjustment to estimate 6 DoF relative position of cameras
• Visualizer to inspect the estimates from the bundle adjustment
• Setting of the World Origin within the visualizer to simplify data processing

3D Tracking

• Tracker API for integrating alternate tracking methods
  • 3 sample implementations with Google Mediapipe (Hands/Pose/Holistic)
• Automated application of landmark tracking to synchronized videos
• Triangulation of 3D landmark position based on calibrated cameras
• Gap-filling and butterworth filtering to smooth trajectory estimates

Trajectory Output

• output to .trc file format for use in biomechanical modelling
• output to tidy .csv format with well-labelled headers for straightforward integration with other workflows
• companion project Rigmarole in development to facilitate creation of animated rigs in Blender

NOTE: Currently only using Mediapipe

Google's Mediapipe provides a relatively easy and efficient method for human subject tracking, though for many uses it is limiting. Caliscope has a general Tracker base class that is implemented in a few versions (Pose/Hands/Holistic). This has provided a proof of concept implementation of markerless tracking, though for more robust use the roadmap calls for integration with more powerful tools such as MMPose and DeepLabCut.

Reporting Issues and Requesting Features

To report a bug or request a feature, please open an issue. Please keep in mind that this is an open-source project supported by volunteer effort, so your patience is appreciated.

General Questions and Conversation

Post any questions in the Discussions section of the repo.

Acknowledgments

This project was inspired by FreeMoCap (FMC), which is spearheaded by Jon Matthis, PhD of the HuMoN Research Lab. The FMC calibration and triangulation system is built upon Anipose, created by Lili Karushchek, PhD. Caliscope was originally envisioned as an alternative calibration tool to Anipose that would allow more granular estimation of intrinsics as well as visual feedback during the calibration process. Several lines of of the original Anipose triangulation code are used in this code base, though otherwise it was written from the ground up. I'm grateful to Dr. Matthis for his time developing FreeMoCap, discussing it with me, pointing out important code considerations, and providing a great deal of information regarding open-source project management.

License

Caliscope is licensed under the permissive BSD 2-Clause license. The triangulation function was adapted from the Anipose code base which is also licensed under the BSD-2 Clause. A primary dependency of this project is PySide6 which provides the GUI front end. PySide6 is licensed under the LGPLv3. Caliscope does not modify the underlying source code of PySide6 which is available via PyPI.