Implementasi Convolutional Neural Network untuk Klasifikasi Rambu-Rambu Lalu Lintas

Andri Firmansyah
Kostaman

SAINTEK
Vol. 4 No. 1 (2025): Seminar Nasional Sains dan Teknologi (SAINTEK) ke 4 - Februari 2025

Traffic Sign Recognition (TSR) technology, which is commonly used for recognizing traffic signs through image processing, can be applied to driver assistance systems, advanced driver assistance systems, autonomous driving systems, road safety, urban environment recognizing, and traffic sign monitoring for maintenance purposes. The dataset used in this study consists of 34,799 traffic sign images categorized into 43 classes. The CNN model consists of two 5x5 convolutional layers, two 3x3 convolutional layers, two 2x2 Maxpool layers, and one fully-connected layer that utilizes the Softmax activation function. The number of filters used in each convolutional layer is 60. By using 22,271 training images, 50 epochs, the obtained error and accuracy values during the training stage are 0.0283 and 99.36%, respectively. In the testing stage with 75 epochs, the CNN model was able to obtain an error value of 0.0400 and an accuracy value of 98.76%.