Viewing corridors as right parallelepipeds for vision-based vehicle localization

Abstract

An approach to vision-based vehicle localization by viewing corridors as a combination of right parallelepipeds is proposed. The objective is to derive the orientation and lateral position of a vehicle in a right parallelepiped corridor, These two kinds of information are all that is needed for vehicles to navigate safely in a right parallelepiped corridor. This approach offers low hardware cost and simple computation. Only one camera mounted on the vehicle is needed, and analytic formulas are derived for computing the vehicle location. The information source is the corridor ceiling. Two orthogonal sets of parallel lines on the corridor ceiling are used to detect the vanishing line of the ceiling. An equation is developed to derive the vehicle orientation by utilizing the detected vanishing line. Also, based on the observation of the variation of image line slopes when we move laterally, another equation is established to evaluate the relative lateral position of the vehicle by utilizing the line slope of the ceiling line pointing forward. Experiments have been conducted, and acceptable vehicle localization results have been obtained, to prove the feasibility of the proposed approach.