Three-dimensional road recognition from single two-dimensional image

A new method to perform three-dimensional road recognition from a single two-dimensional image is proposed. Based on road edges which are characterized by properties of matching points, Euler angles between camera-centered coordinates and edge-centered coordinates can be evaluated and thus the depth information of edge points can be obtained from resultant perspective projections of matching points in the image. In the paper the road segment is assumed as plane and the road width remains invariant, while matching points are defined as edge points with parallel tangent lines which are vertical to the line through the matching points based on the construction of the road. The analysis for the determination of matching points in the image is investigated in detail and a modified algorithm to locate matching points is shown. Experiments are executed on synthetic image data and the results are shown to confirm the validity of the proposed method.