Methods and Software Tools for Converting Road Marking Line Segmentation Mask to Object Representation for Highly Automated Vehicles

This paper describes the development of an algorithm for converting segmentation masks derived from neural networks into ordered sets of point coordinates representing road markings for highly automated vehicles (HAVs). Several methods are considered, including the nearest neighbor method, polynomial approximation, Hough transform, and parallel contouring. Performance tests and complexity analyses are performed for each method. The parallel contour method with dynamic search for opposite points was the most effective, demonstrating linear complexity and high accuracy in approximating road marking contours. This solution ensures efficient processing of large real-time datasets, critical for HAV navigation. The study integrates the algorithm into a unified system, combining segmentation with mask-to-line conversion, enhancing autonomous driving precision.