Toward a new dataset: Mexican Traffic Signs ReWaIn-MTS for detection using deep learning

Abstract

A variety of factors along the road can endanger the safety of drivers or pedestrians and lead to high-impact accidents while driving, which is why traffic signs are essential elements that provide information about the condition of the road during the trip. Traffic sign detection and classification is a research area in computer vision. Its applications are mainly in autonomous conduction or assistance driving. Convolutional neural networks (CNNs) have outstanding detection results compared to conventional methods. In this work, we employed machine learning techniques based on CNNs to categorize and detect Mexican traffic signs. A dataset focused on traffic signs was outlined for the Mexican territory within the main urban roads in eight different cities. The dataset contains 2,283 road elements divided into 37 classes for training and validation of algorithms; a novel methodology is proposed to apply data augmentation and obtain better performance in classification and detection models. The mean Average Precision (mAP) metric compares the performance in state-of-the-art detection methods, particularly YOLOv5, YOLOv8, and the Transformer DETR, obtaining better results with trained models incorporating data augmentation.