pEyeTracker

pEyeTracker is an experimental and cross-platform eye-tracking software written in Python 3. It was designed to primarily work with the Core cameras from Pupil Labs, but it should work with any head-mounted setup with UVC-compliant cameras.

Features

pEyeTracker has some similar functionalities found in older versions of Pupil Capture, but it is definitely not a replacement. If you are looking for a feature-rich eye-tracking software for your UVC cameras, you should stick with Pupil Capture. However, pEyeTracker does have some features that might be particularly interesting for developers:

• virtually no dependency hell
• user interface in QML, decoupled from the eye-tracking code
• architecture is modular and object-oriented
• readable and documented code
• cross-platform (same code runs from source on Windows, Mac, and Linux)
• support for the old 30 Hz binocular Pupil cams

Additionally, pEyeTracker offers some basic functionalities found in similar software (though many others are still missing 😭):

• 2D calibration for each eye
• 3D calibration for each eye
• calibration accuracy measurement
• eye video loading
• gaze data storage
• gaze estimation feedback on interface

Dependencies

• NumPy
• OpenCV 4
• pyuvc
• PySide2
• scikit-learn
• pupil-detectors

Installation and running

pEyeTracker has no packaging (yet), which means that you should run it from source. But first, let's install the dependencies:

pip install -r requirments.txt

Please note that pyuvc is not listed in PyPI, so you may have to install it manually. Look here for further instructions.

Assuming that all went well, access the src folder and start up pEyeTracker with:

python3 main.py

Collaboration

As stated before, pEyeTracker is at an early experimental stage. Unfortunately, I can only work on it at my spare time, so any help with debugging, adding new useful features, or re-writting a problematic module would be greatly appreciated.

If you feel like collaborating, just get in touch or make a pull request with your changes. I will gladly look at it as soon as I can.

Some of the things I think should be implemented at some point in the future:

• a state-of-the-art pupil tracking algorithm (replacing pupil-detectors dependency)
• improved support for 3D eye model and eyeball localization
• provide some useful eye behavior data, such as:
  • number of fixations, saccades, and smooth pursuit events
  • blinks
  • fixation duration
  • saccade length
  • average, min, and max pupil diameter
• support vergence movement detection
• stream gaze data through some network protocol
• option to change calibration parameters on UI
• option to change gaze estimation function and its paramenters on UI
• Unity prefabs to perform calibration with AR/VR devices

Issues

If you need support, just create a new issue. I'll do what I can to help as soon as possible. Please, bear in mind, though, that pEyeTracker is an experimental software with no guarantees or dedicated support.