agrippa-stereocam

Calibration Workflow

Offline stereo calibration is performed in the notebooks under calibration/.

Notebooks

Notebook	Purpose
1.Acquiring_Calibration_Images.ipynb	Live capture via Aravis as an alternative to calibration-capture
2.Calibration.ipynb	Corner detection, stereo calibration, rectification map export
3.Depthmap_with_Tuning_Bar.ipynb	Interactive disparity experimentation in notebook form

Notebook environment setup

cd calibration
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m ipykernel install --user --name agrippa-calibration

Calibration image capture guidance

The quality of the capture set matters more than small solver tweaks.

Use:

• a flat checkerboard with known square size,
• even, diffuse lighting,
• fixed exposure and gain,
• consistent focus for the entire sequence,
• broad coverage of the sensor, including the corners.

Recommended capture count:

• minimum: 15-20 usable stereo pairs
• preferred: 25-40 diverse pairs

2.Calibration.ipynb workflow

The main notebook performs:

1. checkerboard corner detection with subpixel refinement,
2. independent intrinsic calibration for the left and right cameras,
3. stereo calibration using the intrinsic estimates as the initial guess,
4. per-pair reprojection error analysis and outlier rejection,
5. stereo rectification and export of remap tables,
6. epipolar error validation on the rectified pair set,
7. estimation of recommended disparity search parameters,
8. export of calibration_meta.json.

Output

Each calibration session produces a calib_result/ directory containing:

• .npy calibration artifacts,
• remap_left.bin and remap_right.bin for the C runtime,
• calibration_meta.json with summary metadata and quality metrics.

Runtime integration

Those outputs are used by:

• ag-cam-tools capture --calibration-local <session>
• ag-cam-tools stream --calibration-local <session>
• ag-cam-tools depth-preview-classical --calibration-local <session>
• ag-cam-tools depth-preview-neural --calibration-local <session>
• ag-cam-tools calibration-stash upload <session>

You can also store the result on-camera and load it later with --calibration-slot.

This site is open source. Improve this page.