Real-Time and Accurate Pupil Detection Based Retro-Oriented Mind and Ellipse Trend Analysis

Abstract

This study focuses on designing a pupil ellipse detector for wearable eye trackers. The detector uses both a traditional method producing pupil patches inAdifferent resolutions and a learning model segmenting these patches. Therefore, the frequency is increased as the input size of the learning model will be reduced according to the structure of the received image. This novel approach in the pupil detection field was named as Retro-Oriented Mind (ROM). The study also presents metrics measuring the segmentation accuracy and a correction mechanism improving ellipse parameters if metric scores are not acceptable. The combination of novel metrics and correction mechanisms was named as Pupil Ellipse Trend Analysis (PETA). Using ROM and PETA, the proposed study has achieved an accuracy of over 90% and a frequency of more than 120 Hz (from about 30 Hz) in analyses of LPW and Dikablis datasets. These measurements reveal the potential of the study to be used for both medical and general purposes. Code and details: https://github.com/Serif-NNR/rom-peta-pupil-detection.