Traffic Sign Recognition System

Abstract

Traffic sign detection and recognition have become essential components of intelligent transportation systems and Advanced Driver Assistance Systems (ADAS) due to the increasing need for road safety and automated driving. Conventional traffic sign recognition approaches based on handcrafted features and traditional image processing techniques often struggle to achieve high accuracy under varying environmental conditions such as poor lighting, occlusions, motion blur, and complex backgrounds. To overcome these limitations, this work presents a Traffic Sign Detection and Recognition System Using Convolutional Neural Networks (CNN), designed to accurately detect and classify traffic signs from input images. The proposed framework utilizes computer vision techniques for image preprocessing, including resizing, normalization, and image enhancement, followed by deep learning-based feature extraction and classification using a Convolutional Neural Network (CNN). The CNN automatically learns discriminative visual features such as shapes, colors, and patterns from traffic sign images, eliminating the need for manual feature engineering. The system is trained and evaluated using the German Traffic Sign Recognition Benchmark (GTSRB) dataset, which contains more than 50,000 labeled images belonging to 43 different traffic sign classes. A user-friendly interface is developed using Streamlit, enabling users to upload traffic sign images or capture images through a webcam for real-time prediction. The trained model classifies the detected traffic sign and displays the predicted class along with the confidence score. Experimental results are evaluated using Accuracy, Precision, Recall, F1-Score, Confusion Matrix, and Training Performance Metrics, demonstrating the effectiveness of the proposed CNN-based framework for accurate and reliable traffic sign recognition. The developed system contributes to improving road safety and can be effectively integrated into intelligent transportation systems, driver assistance technologies, and autonomous vehicle applications.

Introduction

The text presents a Traffic Sign Detection and Recognition System based on Convolutional Neural Networks (CNNs). The main motivation is to improve road safety by automatically recognizing traffic signs, which can be difficult for drivers and conventional computer-vision systems to identify under challenging conditions such as poor lighting, bad weather, motion blur, occlusion, faded signs, and complex backgrounds.

Traditional traffic sign recognition methods commonly use color thresholding, shape detection, template matching, and handcrafted features. Although these approaches can work in controlled environments, they are sensitive to variations in real-world conditions. More recent machine-learning and deep-learning approaches, including SVM, KNN, Random Forest, CNN, ResNet, YOLO, and EfficientNet, provide better recognition performance. However, some advanced models require substantial computational resources and may be difficult to deploy in lightweight real-time applications.

The proposed system uses a CNN trained on the German Traffic Sign Recognition Benchmark (GTSRB) dataset, which contains more than 50,000 labeled images belonging to 43 traffic-sign categories. CNNs automatically learn important visual features such as edges, colors, textures, shapes, and patterns, eliminating the need for manually engineered features.

Proposed Methodology

The system consists of several major phases:

1. Image Acquisition and Preprocessing
   Images are obtained either through file upload or a live webcam. They are resized, normalized, enhanced, and subjected to noise reduction to improve the quality and consistency of the input.
2. Traffic Sign Detection
   Computer-vision techniques such as color segmentation, edge detection, and contour detection are used to identify potential traffic-sign regions. The relevant Region of Interest (ROI) is extracted and passed to the CNN.
3. CNN-Based Feature Extraction
   Convolutional layers automatically learn visual features from the extracted traffic-sign image. ReLU activation introduces non-linearity, while max-pooling reduces feature dimensions while retaining important information. Fully connected layers then generate a representation suitable for classification.
4. Traffic Sign Classification
   The trained CNN predicts probabilities for the 43 GTSRB classes. The class with the highest probability is selected as the predicted traffic sign.
5. Prediction Display
   A Streamlit-based graphical interface displays the predicted sign and its confidence score. Users can either upload an image or capture one through a webcam, enabling interactive and near-real-time recognition.

Advantages

The proposed system provides several benefits:

• Automatic feature extraction using CNNs
• Recognition of 43 different traffic-sign categories
• Reduced dependence on handcrafted features
• Improved robustness compared with traditional approaches
• Image upload and webcam-based prediction
• Confidence-score-based predictions
• Simple and user-friendly Streamlit interface
• Potential application in Advanced Driver Assistance Systems (ADAS)
• Potential integration into Intelligent Transportation Systems (ITS)
• Possible future use in autonomous-driving applications

Conclusion

This study presented a Traffic Sign Detection and Recognition System using a Convolutional Neural Network (CNN) to provide an intelligent and efficient solution for automatic traffic sign recognition. By combining image preprocessing, computer vision techniques, and deep learning-based feature extraction, the proposed framework accurately detects and classifies traffic signs from uploaded images or live webcam input. The system utilizes the German Traffic Sign Recognition Benchmark (GTSRB) dataset containing 43 traffic sign classes, enabling the model to learn diverse traffic sign patterns and improve recognition performance. The experimental evaluation demonstrates the effectiveness of the proposed framework at different stages of the recognition process. During image preprocessing, techniques such as resizing, normalization, noise reduction, and image enhancement improved the quality of input images and enhanced feature extraction. The CNN model achieved a training accuracy of 97.12% and a validation accuracy of 94.35%, while the training and validation loss decreased steadily throughout the training process, indicating stable model convergence and good generalization. The confusion matrix further confirmed that the model correctly classified the majority of traffic sign categories, with only a few misclassifications occurring among visually similar traffic signs. Additionally, the confidence score distribution showed that most predictions were made with high confidence, demonstrating the reliability of the proposed recognition system. By integrating computer vision with deep learning, the proposed framework provides a significant improvement over traditional traffic sign recognition methods that rely on handcrafted features. The developed Streamlit-based user interface enables users to upload traffic sign images or capture images through a webcam and obtain real-time prediction results along with confidence scores, making the system practical and user-friendly. Overall, the proposed CNN-based Traffic Sign Detection and Recognition System demonstrates high accuracy, robustness, and computational efficiency. The promising experimental results indicate that the framework can effectively support Intelligent Transportation Systems (ITS), Advanced Driver Assistance Systems (ADAS), and future autonomous vehicle technologies by providing accurate and reliable traffic sign recognition under diverse road conditions.

References

[1] G. Singh, A. Kumar, P. Bajaj, Tanisha, and M. Verma, \"Enhancing Road Safety: Convolutional Neural Networks in Traffic Sign Recognition,\" Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2024), IEEE, 2024, pp. 65–72. [2] J. Stallkamp, M. Schlipsing, J. Salmen, and C. Igel, \"Man vs. Computer: Benchmarking Machine Learning Algorithms for Traffic Sign Recognition,\" Neural Networks, vol. 32, pp. 323–332, 2012. [3] K. O\'Shea and R. Nash, \"An Introduction to Convolutional Neural Networks,\" arXiv preprint, 2015. [4] X. Qiao, \"Research on Traffic Sign Recognition Based on CNN Deep Learning Network,\" Procedia Computer Science, vol. 228, pp. 826–837, 2023. [5] L. Alzubaidi, J. Zhang, A. Humaidi, et al., \"Review of Deep Learning: Concepts, CNN Architectures, Challenges, Applications, Future Directions,\" Journal of Big Data, vol. 8, no. 53, 2021. [6] D. Hindarto, \"Enhancing Road Safety with Convolutional Neural Network Traffic Sign Classification,\" Sinkron: Jurnal dan Penelitian Teknik Informatika, vol. 8, no. 4, pp. 2810–2818, 2023. [7] H. Ben Fredj, A. Chabbah, J. Baili, H. Faiedh, and C. Souani, \"An Efficient Implementation of Traffic Signs Recognition System Using CNN,\" Microprocessors and Microsystems, vol. 98, 2023. [8] R. Castruita Rodríguez, C. Mendoza Carlos, O. O. Vergara Villegas, V. G. Cruz Sánchez, and H. de J. Ochoa Domínguez, \"Mexican Traffic Sign Detection and Classification Using Deep Learning,\" Expert Systems with Applications, vol. 202, 2022. [9] J. Cruz Antony, G. M. Karpura Dheepan, V. K., V. Vikas, and V. Satyamitra, \"Traffic Sign Recognition Using CNN and ResNet,\" EAI Endorsed Transactions on Internet of Things, vol. 10, 2024. [10] C. Wang, \"Research and Application of Traffic Sign Detection and Recognition Based on Deep Learning,\" International Conference on Robots & Intelligent System (ICRIS), IEEE, pp. 150–152, 2018. [11] F. Chollet, Deep Learning with Python, 2nd Edition, Manning Publications, 2021. [12] A. Géron, Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow, 3rd Edition, O\'Reilly Media, 2023.

Copyright

Copyright © 2026 Yalla Lokesh Kumar, Dr. T. Siva Ramakrishna. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.