Transfer Learning-Based Speed Limit Traffic Sign Recognition with Multilingual Audio Alerts

In order to maintain road safety and effective traffic management, it is essential to recognize and classify traffic signs. With the rapid increment in the number of vehicles on the road, accurate and efficient speed sign identification systems are in high demand. This research presents a comprehensive ResNet model based on transfer learning for recognizing and classifying speed limit traffic signs with multilingual audio alerts. In India, the need for such a system arises from the country’s diverse linguistic landscape, which requires effective communication of speed limit signs to drivers in their preferred language. We used nine-speed limit sign classes each from the German Traffic Sign Recognition Benchmark (GTSRB) and Indian Traffic Signs Prediction (85 classes) datasets. Our trained models achieved impressive test accuracies of 99.96 and 98.62% using ResNet50, among other state-of-the-art methods, including CNN and VGG. Recognizing drivers’ proclivity to ignore speed limit signs, our initiative aims to improve road safety by informing drivers of speed limits by delivering audio alerts in their preferred Indian language.