The authors describe a road-following system for an autonomous land vehicle, based on range image analysis. The system is divided into two parts: low-level data-driven analysis, followed by high-level model-directed search. The sequence of steps performed in order to detect three-dimensional (3-D) road boundaries is as follows. Range data are first converted from spherical into Cartesian coordinates. A quadric (or planar) surface is then fitted to the neighborhood of each range pixel, using a least squires fit method. Based on this fit, minimum and maximum principal surface curvatures are computed at each point to detect edges. Next, using Hough transform techniques, 3-D local line segments are extracted. Finally, model-directed reasoning is applied to detect the road boundaries.>

This paper describes a model-based geometric reasoning module for autonomous road following. Vision-guided road following requires extracting road boundaries from images in real-time to guide the navigation of autonomous vehicles on roadway. The detected road region boundary is error-prone due to imperfect image segmentation. To achieve robust system performance, a geometric reasoning module that uses spatial and temporal constraints to perform model-based reasoning is required. Local geometric supports for each road edge segment are collected and recorded and global consistency checking is performed to obtain a consistent interpretation of the raw data. Cases involving incomplete sensor data, on curved roads where only one side of the road is visible, and incorrect segmentation due to shadows, road patches, or unusual road conditions, can usually be detected and corrected. This reasoning module has been integrated into a road following system which is capable of supporting autonomous road following at 19 km/hr.