Estimating Self-Motion with a Spherical Camera versus Pinhole Camera

Taking inspiration from the eyeball we propose a spherical camera model for the estimation of optic flow and self-motion. Our contribution is the definition of a spherical optic flow model using symmetric spherical coordinates, which allows for the formulation of a linear method estimating self-motion from optic flow. We compare our method with a method that estimates self-motion from optic flow for a pinhole camera. Estimates from both camera models are comparable while varying the zoom, spatial sampling, field of view, or temporal sampling. Selecting 80% of the best optic flow estimates gave better estimates than taking all optic flow for both camera models. For same parameter choices of zoom, sampling, or field of view the spherical camera model gave better estimates of optic flow and self-motion than the pinhole camera model. For the task of estimating self-motion a spherical camera performs no worse than a pinhole camera and provides a wider field of view – an advantage not only to the estimation of self-motion but also for tasks such as visual search, obstacle avoidance, or visual navigation.