Variations of pupil centration and their effects on video eye tracking

To evaluate measurement errors that are introduced in video eye tracking when pupil centration changes with pupil size.

Software was developed under Visual C++ to track both pupil centre and corneal centre at 87 Hz sampling rate at baseline pupil sizes of 4.75 mm (800 lux room illuminance) and while pupil constrictions were elicited by a flashlight. Corneal centres were determined by a circle fit through the pixels detected at the corneal margin by an edge detection algorithm. Standard deviations for repeated measurements were ± 0.04 mm for horizontal pupil centre position and ± 0.04 mm for horizontal corneal centre positions and ±0.03 mm for vertical pupil centre position and ± 0.05 mm for vertical corneal centre position. Ten subjects were tested (five female, five male, age 25-58 years).

At 4 mm pupil sizes, the pupils were nasally decentred relative to the corneal centre by 0.18 ± 0.19 mm in the right eyes and -0.14 ± 0.22 mm in the left eyes. Vertical decentrations were 0.30 ± 0.30 mm and 0.27 ± 0.29 mm, respectively, always in a superior direction. At baseline pupil sizes (the natural pupil sizes at 800 lux) of 4.75 ± 0.52 mm, the decentrations became less (right and left eyes: horizontal 0.17 ± 0.20 mm and -0.12 ± 0.22 mm, and vertical 0.26 ± 0.28 mm and 0.20 ± 0.25 mm). While pupil decentration changed minimally in eight of the subjects, it shifted considerably in two others. Averaged over all subjects, the shift of the pupil centre position per millimetre pupil constriction was not significant (right and left eyes: -0.03 ± 0.07 mm and 0.03 ± 0.04 mm nasally per mm pupil size change, respectively, and -0.04 ± 0.06 mm and -0.05 ± 0.12 mm superiorly). Direction and magnitude of the changes in pupil centration could not be predicted from the initial decentration at baseline pupil sizes.

In line with data in the literature, the pupil centre was significantly decentred relative to the corneal centre in the nasal and superior direction. Pupil decentration changed significantly with pupil size by 0.05 mm on average for 1 mm of constriction. Assuming a Hirschberg ratio of 12° mm(-1) , a shift of 0.05 mm is equivalent to a measurement error in a Purkinje image-based eye tracker of 0.6°. However, the induced measurement error could also exceed 1.5° in some subjects for only a 1 mm change in pupil size.