Intelligent Road Sign Detection and Categorization Framework

Research on an AI-enabled road sign detection system has been ongoing because of its numerous real-world uses. Many of the existing solutions, which frequently rely on numerous constraints, are still not reliable in practical settings. A reliable and effective road sign detection and classification system built on the cutting-edge YOLO object detector is presented in this study. Each detection stage involves training and fine-tuning the Convolutional Neural Networks (CNNs) to make them resilient to various situations, such as changes in the camera, backdrop, and lighting. We specifically designed a two-step method for object identification, using a basic YOLOV8 model to recognize road-sign symbols in real-time. In the second stage, we convert the model’s output into audio format. In one dataset, the resultant method produced remarkable outcomes. The dataset included 2093 frames from 20 distinct classes, and our system generated 59 frames per second (FPS) with a recognition rate of 94.9%. This was a significant improvement over the prior results (81.80%) and surpassed the commercial systems SqueezeNet and MobileNet (89.80% and 93.03%, respectively) as well as FasterRCNN. The commercial systems’ experimental versions achieved identification rates below 70% in our suggested dataset. However, with a precision of 97.4 and 35 FPS, our system outperformed.