A zero-bank algorithm for inverse perspective of a road from a single image

A method is presented for reconstructing the 3-D geometry of a road from a single image of the road. This problem has an infinity of solutions unless restrictive hypotheses about geometric characteristics of this road are assumed. The road is modeled as a space ribbon defined by a spine (centerline) and generators (cross-segments) which are horizontal line segments cutting the spine at their midpoint at a normal angle. Properties of two neighboring generators of such a ribbon are examined; if a generator is known, a neighboring generator is completely defined if one of its ends is known. The proposed method uses this property to reconstruct the visible part of the world road, by iteratively finding a series of generators. This method is tested against a simple method which assumes that the ground is flat ("Flat Earth assumption"), and against another method which uses vanishing points.

[1]  Larry S. Davis,et al.  Road Boundary Detection For Autonomous Vehicle Navigation , 1986 .

[2]  Allen M. Waxman,et al.  Visual navigation of roadways , 1985, Proceedings. 1985 IEEE International Conference on Robotics and Automation.

[3]  Azriel Rosenfeld,et al.  Axial representations of shape , 1986, Computer Vision Graphics and Image Processing.

[4]  Azriel Rosenfeld,et al.  Synthesis of a road image as seen from a vehicle , 1986, Pattern Recognit..

[5]  Ramesh Jain,et al.  Road Following Using Vanishing Point , 1987 .

[6]  Shih-Ping Liou,et al.  Detecting road edges using hypothesized vanishing points , 1985 .