Year: 2025
Type: article
Abstract: Recognition of traffic signs is one of the key activities in the development of autonomous vehicles for safe navigation on the roads. This work addresses the study of ConvNet in classifying Turkish traffic signs into two classes: hazard-warning signs and regulatory signs. A dataset of 129 traffic sign images was utilized, augmented through hue jitter transformations to enhance model performance. The ConvNet, based on a three-convolution-layer architecture, four ReLU layers, and two fully connected layers, is trained to classify the two classes of traffic signs. The attained average accuracy was 97.7% ± 5.2% on the training set, 88.8% ± 1.2% on the validation set, and 96.9% ± 7.2% on the test set. These results really prove that ConvNets work quite well in identifying and classifying traffic signs, thus proving that they can be applied in autonomous vehicle technologies. Real-world photos of traffic signs will be used in future studies to test the model's applicability.
Authors: Mehmet Özcan, Abdurrahim Hüseyin Ezirmik
Institution: Balıkesir University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4415436439