Open pandacriss opened 6 years ago
I'm not working on this project anymore. Someone could write this in a fork though and link it here.
In case if somebody is looking for the same: I found it already done here: https://github.com/abhisuri97/pyeLike
I have tried some other similar projects based on the same idea and paper: https://github.com/noelledavis/gaze_tracking https://github.com/jonnedtc/PupilDetector however, they are slow and very often not very accurate.
So, eyeLike (and pyeLike) are the most accurate and fast solutions. I fed a bunch of test images and there were some failures, but I can deal with them applying face_recognition
library https://github.com/ageitgey/face_recognition (wrapper around dlib) to extract eye region and then to reduce it even more to leave only inner area of the eye, and also applying img = cv2.equalizeHist(img)
(but do it only after extracting the inner region of the eye, otherwise it can make eyeLike work worse).
could you add a way to call the find eyes function from python? tat would be very helpfull