Off-Line Vehicle Localization through a P3P Solution by Omnidirectional Spherical Image

Abstract

We propose a strategy to estimate the self position and pose of a vehicle by using spherical images. Our method is based on the Perspective-Three-Point (P3P) problem from three pairs of 2D-3D correspondences. Our method introduces a quadratic form of a vector with a 3-rank tensor, which can be solved easily and quickly by using a Newton-Raphson method. A utilization of a spherical image in place of a conventional image overcomes the lack of accuracy for the localization. We present some experimental results that show the practicality and high precision of our method by using real data sets.