Perspective Pose Estimation from Uncertain Omnidirectional Image Data

Omnidirectional vision is highly beneficial for robot navigation. We present a novel perspective pose estimation for omnidirectional vision involving a parabolic central catadioptric sensor using small data sets. We incorporate an appropriate and approved stochastic method to deal with uncertainties in the data. Our approach is robust in that it is more accurate than recent methods while using less precise hardware without rigorous calibration