Robust Real-Time Lane Delimiting Features Extraction

This paper presents a robust lane marking detection algorithm suited for a large variety of traffic scenarios, under various lighting conditions. The algorithm detects dark-light-dark transitions using an improved method for horizontal gradient computation, which accounts for the expected marking width and for the perspective effect by means of a variable size filter. The computation of the variable filter-based gradient is optimized for real-time operation by reusing the intermediate results of the neighboring pixels. Stereovision is used for final filtering of the results, by selecting only the marking-like features located on the road surface.