METHODS AND ALGORITHMS FOR SEGREGATING TRAFFIC SIGNS BY SEGMENTATION USING NEURAL NETWORKS

This article investigates neural-network-based methods and algorithms for extracting road traffic signs via image segmentation. The primary objective is to build a robust real-time solution capable of stable performance under varying illumination, complex noise, and high background clutter. The methodology systematically evaluates U-Net and Mask R-CNN family architectures enhanced with domain-specific data augmentation techniques and tailored post-processing filters. As a core scientific novelty, a multi-stage segmentation pipeline coupled with an adaptive loss formulation is analyzed to improve boundary compliance and drastically minimize false positives. Quantitative results prove that incorporating boundary-sensitive constraints provides superior edge precision for downstream autonomous driving tasks.