Traffic signs recognition for advanced driver assistance systems using deep learning

Traffic accidents remain a major global cause of death, motivating the integration of artificial intelligence (AI) with advanced driver assistance systems (ADAS). Within ADAS, traffic sign recognition (TSR) is essential for interpreting road environments and improving driving safety. However, challenges such as changes in lighting, poor weather conditions, and blocked signs continue to limit recognition accuracy. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have substantially improved TSR performance. This study evaluates deep learning architectures for TSR on the German Traffic Sign Recognition Benchmark (GTSRB) dataset across three research pathways. First, a benchmarking study reproduces the Custom CNN and VGG16 architectures proposed by the existing paper under a closely matched experimental protocol, with the reproduced Custom CNN achieving a validation accuracy of 97.82%, surpassing the 94–96% reported in the baseline. Second, three newly introduced architectures AlexNet, Xception, and YOLOv11 are evaluated, with Xception achieving 92.95% validation accuracy and YOLOv11 achieving a test mAP50 of 97%, suggesting a potential advantage of detection-based frameworks for real-time ADAS applications in this single-run, single-dataset evaluation. Third, two transformer-based architectures ViT-B/16 and DeiT-Tiny are developed and evaluated, achieving test accuracies of 96.96% and 96.61% respectively, with DeiT-Tiny delivering superior precision, recall, and F1-score despite being 15 times smaller. Results demonstrate that transformer-based and detection-based models offer compelling advantages over traditional CNN baselines, with DeiT-Tiny emerging as the most parameter-efficient architecture and YOLOv11 providing the strongest real-time spatial localization capability for ADAS deployment.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-21
DOI
https://doi.org/10.1038/s41598-026-75204-w
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Traffic signs recognition for advanced driver assistance systems using deep learning

Mahmoud A. Shawky, Tamer Shamseldin, Mona Hesham, Mohammed Abdrabou
Scientific Reports
Advanced Neural Network Applications
article

Traffic signs recognition for advanced driver assistance systems using deep learning

Mahmoud A. Shawky, Tamer Shamseldin, Mona Hesham, Mohammed Abdrabou
article en

Abstract

Traffic accidents remain a major global cause of death, motivating the integration of artificial intelligence (AI) with advanced driver assistance systems (ADAS). Within ADAS, traffic sign recognition (TSR) is essential for interpreting road environments and improving driving safety. However, challenges such as changes in lighting, poor weather conditions, and blocked signs continue to limit recognition accuracy. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have substantially improved TSR performance. This study evaluates deep learning architectures for TSR on the German Traffic Sign Recognition Benchmark (GTSRB) dataset across three research pathways. First, a benchmarking study reproduces the Custom CNN and VGG16 architectures proposed by the existing paper under a closely matched experimental protocol, with the reproduced Custom CNN achieving a validation accuracy of 97.82%, surpassing the 94–96% reported in the baseline. Second, three newly introduced architectures AlexNet, Xception, and YOLOv11 are evaluated, with Xception achieving 92.95% validation accuracy and YOLOv11 achieving a test mAP50 of 97%, suggesting a potential advantage of detection-based frameworks for real-time ADAS applications in this single-run, single-dataset evaluation. Third, two transformer-based architectures ViT-B/16 and DeiT-Tiny are developed and evaluated, achieving test accuracies of 96.96% and 96.61% respectively, with DeiT-Tiny delivering superior precision, recall, and F1-score despite being 15 times smaller. Results demonstrate that transformer-based and detection-based models offer compelling advantages over traditional CNN baselines, with DeiT-Tiny emerging as the most parameter-efficient architecture and YOLOv11 providing the strongest real-time spatial localization capability for ADAS deployment.

Scientific ReportsVol. 16(1)
British University in Egypt (EG)
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Traffic signs recognition for advanced driver assistance systems using deep learning — Mahmoud A. Shawky, Tamer Shamseldin, et al. · Scientific Reports (2026) | TGRS Research Map | TGRS