Road feature detection and estimation

Abstract.

Crosswalks and stair-cases are useful road features for outdoor navigation. In this paper, crosswalks and stair-cases are detected by looking for groups of concurrent lines, and edges are then partitioned using intensity variation information. To distinguish them, three methods are developed to estimate the pose: a homography search approach using an a priori model, and finding the normal using the vanishing line computed from equally spaced lines and likewise with two vanishing points. These algorithms, with error analysis carried out, have been applied to real images with promising results, and they are also useful in other shape-from-texture applications.