Estimation of omnidirectional camera model with one parametric projection

Abstract

This paper presents a new self-calibration algorithm of omnidirectional camera from uncalibrated images. First, one parametric non-linear projection model of omnidirectional camera is estimated with the known rotation and translation parameters. After deriving projection model, we can compute an essential matrix of unknown camera motions, and then determine the camera positions. In addition, we showed that LMS (Least-Median-Squares) is most suitable for inlier sampling in our model than other methods: 8-points algorithm and RANSAC (RANdom Sampling Consensus). In the simulation results, we demonstrated that the proposed algorithm can achieve a precise estimation of the omnidirectional model and extrinsic parameters. © Springer-Verlag Berlin/Heidelberg 2006.