Fisheye omnidirectional camera calibration — Pinhole or spherical model?

In this paper, we investigate two camera models for fisheye camera, namely, the pinhole camera model and spherical projection model. The former one has been successfully applied in calibration modeling of monocular vision system. However, for fisheye camera calibration modeling, we demonstrate that the spherical projection model is more appreciate for nonlinear image system. Moreover, to deal with the high distortion introduced by the camera lens, we use local Hough Transform to detect sub-pixel corners in stead of some classical corner detectors, and two optimization methods are proposed to estimate the camera parameters. Experimental results validate the approach.