To understand human vision it is important to appreciate the challenges faced by the brain in acquiring stable images of natural three dimensional (3D) scenes and controlling eye movements. Recent advances in computer simulation now make it possible to create realistic models of these essential peripheral processes of human vision. Specifically, we describe physically based models for simulating eye movements controlled by extraocular muscles (EOMs) and image formation on a curved retina.
We simulate 3D eye movements using a new biomechanical simulation framework. Most previous models simplify the nonlinearities of the oculomotor plant's geometry and EOM mechanics, and are limited to static simulation. We model EOMs as a collection of “strands,” which are modeling elements for musculotendon dynamics based on splines with inertia. Each strand represents a part of the musculotendon aligned with the fibers; depending on the level of detail needed, a strand can be as fine as a single fascicle or as coarse as an entire muscle. Anatomical variations in EOM and globe geometry across individuals can be taken into account. The resulting model generates realistic gaze positions and trajectories from brainstem control signals.
We can simulate image formation on the retina by tracing a large number of rays from 3D objects as they pass through the optics of the eye to form an image on the curved retina. Most existing models use a pin-hole camera model, with images formed on a planar surface. A more realistic model is useful for understanding basic questions of vision science such as binocular alignment and transsaccadic perception. Recent advances in computer graphics hardware make it possible to efficiently compute these retinal images at interactive rates.
NIH grant 1R01NS050942, Peter Wall Institute for Advanced Studies, NSERC, and Canada Research Chairs Program.