Calibration Toolkit

This repository contains small scripts for camera calibration and for inspecting/editing mocap point files used during extrinsic calibration.

Folder structure (common)

Many scripts assume a “dataset root” folder that contains at least:

<dataset_root>/
├── mocap_points/
│   ├── <name>.txt
│   └── ...
└── extrinsic_images/
      ├── <name>.jpg (or .png/.jpeg/...)
      └── ...

Notes:

• Each mocap_points/<name>.txt is whitespace-separated x y z per line.
• Image matching is done by filename stem (<name>).

Live mocap viewer/editor

Script: plot_mocap_points.py

Purpose:

• Live 3D viewer for a mocap .txt file (points are labeled with their row index).
• Auto-refreshes when the currently viewed file changes on disk.
• Lets you navigate adjacent numbered files and apply in-place transforms.
• Shows a preview of the matching image from extrinsic_images/.

How to run:

python plot_mocap_points.py

Initial configuration (top of file):

• file_path: initial mocap file to open.
• adjust_sector_size: the “sector size” used by the reverse operation (default 8).
• sort_size: (rows, cols) grid size used for the flip logic (default (6, 8)).

UI controls:

• Prev / Next: switches _N index in the filename (e.g., Foo_4.txt → Foo_5.txt).
• Reverse sectors (...): reverses the order within each sector (every adjust_sector_size lines), writing back to the same file.
• Sort file: sorts points and writes back to the same file.
• dataset root + Browse: change to another dataset folder containing mocap_points/ (+ extrinsic_images/ for preview).

Sorting controls (right panel):

• perspective: auto/left/right/center. auto infers from folder names (e.g., left, right, middle, center).
• x_flip / fb_flip: optional grid flips after sorting.
  • x_flip = reverse each row (left/right inversion).
  • fb_flip = reverse row order (front/back inversion).
• rows / cols: grid dimensions used for flips.
• u-axis / v-axis: optional axis override for sorting (none, x, -x, y, -y, z, -z).
  • Sorting rule: sort by v descending (top→down), then by u ascending (left→right).

Important:

• Reverse sectors and Sort file modify the viewed .txt in-place. Keep a copy if you need to preserve the original.
• Browse uses tkinter. If it’s unavailable, paste the dataset path into the dataset root box and press Enter.

Camera calibration (intrinsic + extrinsic)

Script: main_cali_all.py

What it does:

• Intrinsic calibration using chessboard images (OpenCV) and writes intrinsic.json.
• Extrinsic calibration using solvePnP per image and writes extrinsics.json (also includes reprojection errors and best_extrinsic).

Before running:

• Edit the constants near the top of main_cali_all.py:
  • chessboard_rows, chessboard_cols, chessboard_length_mm
  • input_folder (your dataset root)

Expected inputs (per current code):

• Images: <input_folder>/extrinsic_images/*.jpg
• Mocap points: <input_folder>/mocap_points/sorted/<image_stem>.txt

Run:

python main_cali_all.py

Outputs:

• <input_folder>/intrinsic.json
• <input_folder>/extrinsics.json
• <input_folder>/ChessboardCorners/ (if enabled)
• <input_folder>/undistorted_images/

Capture images from a webcam

Script: cap.py

• Opens webcam index 0, shows a live view.
• Press Q to save a frame into calibration_images/.
• Press Esc to quit.

Rotate images in-place

Script: rotate_images.py

• Rotates all .jpg images in a configured folder (in-place).
• Edit folder_path and degrees_to_rotate in the script.

3D→2D projection demo (mocap → video)

Script: map_mocap_points.py

• Demonstration code that:
  • parses an OptiTrack CSV,
  • maps markers to a Human3.6M-like ordering,
  • loads intrinsic.json + extrinsics.json,
  • projects 3D points to 2D using OpenCV and overlays them on a video.

It is a demo script: edit the file paths in __main__ to your local data.

Dependencies

Minimum commonly used packages:

pip install numpy pandas matplotlib opencv-python

Optional (only for rotate_images.py):

pip install pillow