Abstract
The illusion of depth generated by linear perspective is very tolerant to the viewing position. Movies and paintings generally look fine from almost anywhere in the room. It is not necessary to view the image exactly from the point through which the 3D scene was projected onto the image plane. However, a tight match between the projection point and the viewpoint is required in the application of augmented reality, which presents users a view of their surroundings that is superimposed with stereoscopic imagery. If the projection and view points don't match, the virtual content won't align well with the real world. We asked: where in the human eye is the viewpoint? A survey of the vision science literature suggests that it is commonly assumed to lie near the nodal points of the eye. The nodal points, a concept from paraxial optics, lie on the optical axis and are defined by the property that the angle of a ray of light reaching the front nodal point is identical to the angle of the ray leaving the back nodal point. However, when we replicated the paraxial analysis and included an aperture stop to represent the iris of the eye, we found that it is more accurate to consider the viewpoint to lie near the entrance pupil. The entrance pupil is the location of the image of the hole in the aperture stop as seen from the front of the optical system. Our subsequent ray tracing analysis using established schematic eyeball models (Arizona, Navarro, Gullstrand, and others) confirmed that projection through the entrance pupil gives more veridical results than projection through the nodal points. We discuss the implications of moving the viewpoint from the nodal points to the entrance pupil in the context of augmented reality, as well as stereopsis and monocular parallax.