Images are a powerful and important means of communication in computer graphics and visualization, but they also serve as a quality approximation of complex 3-D scene geometry. Most images are produced using the planar pinhole camera model (PPC), because of its familiarity to users and simplicity of implementation. There are three key limitations to the planar pinhole camera model:
Planar pinhole cameras are limited to a 180° field of view preventing panoramic views of the scene.
The field of view is sampled in a spacially uniform manner, precluding sampling based upon importance.
The planar pinhole camera only samples what is visible from the pinhole preventing sampling when direct line-of-sight is not available.
In order to create images that better sample 3-D scenes, I propose abandoning the constraints of the conventional planar pinhole camera model by no longer requiring that rays be straight, converge, or sample space uniformly. Camera models can then be designed for specific applications and optimized dynamically for each 3-D scene or dataset so as to achieve adequate sampling. At the same time, camera models should also be designed to preserve image computation efficiency in order to support interactive rendering of dynamic scenes.
Paul Rosen, “Improved 3-D Scene Sampling by Camera Model Design”, Ph.D. Dissertation, Purdue University, July 2010.
