On the appropriateness of camera models

Abstract

Distinct camera models for the computation of structure from motion (sfm) can be arranged in a hierarchy of uncalibrated camera models. Degeneracies mean that the selection of the most appropriate camera from the hierarchy is key. We show how accuracy of fit to the data; efficiency of computation; and clarity of interpretation enable us to compute a measure of the appropriateness of a particular camera model for an observed trajectory set and thus automatically select the most appropriate model from our hierarchy of camera models. An elaboration of the idea, that we call the combined appropriateness allows us to determine a suitable frame at which to switch between camera models.