Traffic Sign Board Recognition Using Computational Techniques

Abstract

Abstract Traffic signs are essential for traffic management, regulating driving conduct, and lowering accidents, injuries, and fatalities. Any Intelligent Transportation System must have automatic traffic sign detection and identification. This work describes a deep-learning-based autonomous technique for traffic sign recognition in India. End-to-end learning with a Convolutional Neural Network (CNN) and a Support Vector Machine (SVM) provided inspiration for automatic traffic sign detection and recognition. The proposed concept was evaluated using a cutting-edge dataset consisting of 12933 images with divided into training set and testing set. In training set total of 8422 images of 10 categories and testing Data consists of 4511 images with 10 categories representing multiple instances of Indian traffic signs. Real-time photographs are taken on metropolitan city roads for the testing dataset and training dataset. By using Google Collab, the suggested system starts by detecting traffic signs in photos using CNNs, which are well-suited to feature extraction in visual data. After detecting the indications, a mixture of CNN and SVM classifiers is used to accurately recognize the indicators. The suggested technique achieves excellent accuracy and real-time performance, as demonstrated by experimental findings on Realtime traffic sign recognition datasets. After analyzing of the training dataset and testing dataset the results we achieved with a test accuracy of 1.0 for the proposed CNN model and test accuracy for the SVM model is 0.84 which is lower than the CNN model. This study advances traffic sign recognition systems, which are critical for increasing road safety, navigation, and allowing autonomous cars to make educated judgments in real-world traffic conditions. The combination of CNNs and SVMs offers a potential method for accurate and dependable traffic sign identification, solving the problems provided by complex and dynamic traffic scenarios.