|
Development of a Web-Based Application for Automatic Traffic Sign Detection and Recognition Using Teachable Machine |
|---|---|
| รหัสดีโอไอ | |
| Creator | Thunthanapon Pirom |
| Title | Development of a Web-Based Application for Automatic Traffic Sign Detection and Recognition Using Teachable Machine |
| Contributor | Pawarisa Hamniam, Warattapop Thapatsuwan, Peeraya Thapatsuwan |
| Publisher | Department of Computer Education, Faculty of Science and Technology, Surindra Rajabhat University |
| Publication Year | 2568 |
| Journal Title | Journal of Computer and Creative Technology |
| Journal Vol. | 3 |
| Journal No. | 2 |
| Page no. | 169-184 |
| Keyword | Traffic Sign Recognition, Teachable Machine, Convolutional Neural Networks, Data Augmentation, Web-based Image Classification |
| URL Website | https://so13.tci-thaijo.org/index.php/jcct |
| Website title | Journal of Computer and Creative Technology |
| ISSN | ISSN 2985-1580 (Print);ISSN 2985-1599 (Online) |
| Abstract | This research aims to develop a model for automatic traffic sign classification using Teachable Machine, a machine learning platform that requires no programming knowledge. The model employs Convolutional Neural Networks, specifically MobileNet architecture, trained using the Adam Optimizer on TensorFlow.js, enabling client-side deployment via a web application. The dataset used consists of 17,440 traffic sign images across 27 classes, compiled from the standard GTSRB dataset and additional photographs taken in Thailand. Data Augmentation techniques, including image rotation, lighting adjustment, noise injection, and perspective transformation, were applied to enhance data diversity and reduce overfitting risk. The model was trained with the following parameters: batch size of 16, 50 training epochs, and a learning rate of 0.001. Model performance was evaluated using standard metrics: Accuracy, Precision, Recall, and F1-Measure, calculated from a 27 × 27 Confusion Matrix. The learning behavior was further analyzed through Accuracy per Epoch and Loss per Epoch graphs. Experimental results indicate that the model achieved high performance, with an average accuracy of 98.93% and an F1-Measure of 98.90%, demonstrating the potential of Teachable Machine in developing accurate models applicable to both educational contexts and intelligent traffic systems. |
