Key points are not available for this paper at this time.
This paper presents an approach to visually classifying the high-level geometry of the road ahead of a vehicle as a section with continuous curvature parallel edges or an intersection containing right-angled legs. The default behavior of the system is snake-based tracking of parallel road edges for curvature estimation. A separate process segments the road surface from the background using color appearance characteristics, then classifies the segmented road shape in a rectified view as either a standard section or one of several intersection types ("four-way", "T", "right angle", etc.). Recognition of the proper shape class of the approaching road is a prerequisite for switching between shape templates in order to successfully track road edges as the vehicle travels through intersections.
Christopher Rasmussen (Mon,) studied this question.