TRAFFIC SIGN RECOGNITION WITH YOLOv5

Abstract

Nowadays, road traffic accidents are increasing dramatically, and one of the most reasons is the lack of knowledge about the traffic signs of the drivers. Traffic sign recognition has long been an important component of driver-assistance systems, as it may help drivers avoid a wide range of possible hazards while also improving their driving experience. On the other hand, traffic sign recognition is a realistic assignment with numerous constraints, like the visual environment, physical damages, and partial occasions. Therefore, it is necessary to have an application that can meet these requirements in order to reduce the number of traffic accidents because of the lack of understanding of traffic signs. In this study, the object detection algorithm - YOLOv5, a unified deep learning architecture for real-time recognition applications, is used for recognizing traffic signs due to its speed and accuracy. The traffic signs dataset used for this study was modified from the Zalo AI Challenge 2020 including 4469 images belonging to 7 different classes. The result shows that the model has achieved good performance with the recognition of traffic signs, even with the small ones.