| The estimation of eyelid position relative to the pupil is important for improving the accuracy of eye tracking systems and estimating subject fatigue level. Novel algorithms were developed for the detection of eyelid relative to the pupil by a band of local minima appearing at the boundary of the eyelid and the iris, and by the change in tangential angle of the pupil boundary. The algorithms were optimized to recognize all images with partial pupil coverage, resulting in a false negative rate of 9.5%. Another algorithm was developed to measure the percentage of pupil coverage, which was shown to have a RMS error of 4.2%. The algorithms were used successfully to reduce the bias in pupil centre estimates due to eyelid interference to within 1° of visual angle of the expected pupil centre position when the pupil was covered by less than 30%. |
