Calibrate gaze position

Gaze position calibration is an optional ArFrame pipeline step. It processes each new gaze position before any further pipeline steps.

The calibration algorithm can be selected by instantiating a particular GazePositionCalibrator from the GazeAnalysis submodule or from another Python package.

Enable ArFrame calibration

Gaze position calibration can be enabled with a dedicated JSON entry.

Here is an extract from the JSON ArFrame configuration file where a Linear Regression calibration algorithm is selected with no parameters:

{
    "argaze.ArFeatures.ArFrame": {
        "name": "My FullHD screen",
        "size": [1920, 1080],
        ...
        "gaze_position_calibrator": {
            "LinearRegression": {}
        },
        ...
    }
}

Edit calibration parameters

# Assuming the ArFrame is loaded
...

    # Start calibration process
    ar_frame.gaze_position_calibrator.reset()

    # Assuming that expected and observed gaze positions are available
    ...

        # If calibration process started
        if ar_frame.gaze_position_calibrator.is_calibrating():

            # Store calibration data
            ar_frame.gaze_position_calibrator.store(timestamp, observed_gaze_position, expected_gaze_position)

    # End calibration process
    score = ar_frame.gaze_position_calibrator.calibrate()

Save calibration parameters as JSON file

Calibration parameters can be save as JSON file:

ar_frame.gaze_position_calibrator.to_json('calibration.json')

Here is the saved JSON file where calibration parameters are stored:

{
    "argaze.GazeAnalysis.LinearRegression": {
        "coefficients": [
            [
                0.901167941442693,
                0.0345129853595345
            ],
            [
                0.11551395622739168,
                0.9315744785596141
            ]
        ],
        "intercept": [
            65.43372920399452,
            -52.23141937917768
        ]
    }
}

Load calibration parameters file

Saved calibration parameters can be loaded from JSON ArFrame configuration file:

{
    "name": "My FullHD screen",
    "size": [1920, 1080],
    ...
    "gaze_position_calibrator": "calibration.json"
    ...

They also can be loaded from Python script:

ar_frame.gaze_position_calibrator.from_json('calibration.json')