Abstract

The stringent requirements of achieving both high spatial (0.25°) and temporal (1-ms) resolution in a helmet-mounted electrooptical system permitting 2-D gaze measurements independent of head translation prohibits implementation of 2-D sensors. To meet these requirements, SRD has recently introduced a novel technique wherein two 1-D, photoelectric self-scanned arrays detect projections of the bright pupil (P) and corneal reflection (CR) images, thereby reducing the number of sampled pixels from n² to 2n. To achieve subpixel resolution we have developed algorithms based on models of the P, CR, and iris signals. The pupil center is immediately provided with subpixel resolution once its elliptic model parameters are determined by a least-squares fit of the P-set of data. Alternatively it can be determined using the P-edges defined by the intersection of the P-model with the iris model. The P-model enables definition of its center and/or edges even in extreme cases where the CR obscures one of the latter. The distorted CR signal is first restored by subtraction of the P-model from the raw data. The set of 5–7 points of the restored CR signal is then used for computation of a Gaussian model to provide the center of the CR with 0.25 pixel resolution. Measurements with an artificial eye indicate resolution (and accuracy) of better than 0.25° at a rate of 1000 frames/s.

© 1986 Optical Society of America