Driver Assistance through Real-Time Traffic Sign Recognition Using HSV Color Space and Lightweight CNN Models

This paper presents a real-time traffic sign recognition (TSR) system designed to improve driver safety by accurately detecting and classifying road signs under varying illumination and occlusion conditions. The system combines classical image processing (HSV color segmentation and shape-based feature extraction) with a lightweight convolutional neural network (CNN) classifier. Robustness is increased through an extensive data-augmentation pipeline (geometric transforms, photometric changes, and synthetic occlusions). For deployment we describe a low-power embedded hardware option and acceleration strategies to meet real-time constraints. Experimental evaluation on the German Traffic Sign Recognition Benchmark (GTSRB) and additional in-house test sets demonstrates strong classification accuracy (> 98% on clean data) with latency suitable for in-vehicle operation (target ≥ 15 fps). The paper details methodology, implementation, results, and design trade-offs, providing a practical blueprint for production-ready TSR systems.