Title: Efficient traffic sign recognition with YOLOv5

Authors: Omran Nacir; Maraoui Amna; Werda Imen; Belgacem Hamdi

Addresses: LEμE, University of Monastir, Monastir, Tunisia ' LEμE, University of Monastir, Monastir, Tunisia; LEAT, University Cote d'Azur, Cote d'Azur, France ' LETI, University of Sfax, Sfax, Tunisia ' LEμE, University of Monastir, Monastir, Tunisia

Abstract: The successful integration of autonomous vehicles into urban environments requires strict adherence to traffic rules and regulations. A critical component for achieving this level of compliance is an efficient and reliable vision system that can accurately recognise and detect traffic signs. In this paper, we propose a comprehensive vision system built upon trained YOLOv5 models for traffic sign classification and detection. Our vision system utilises the YOLOv5 algorithm, which has been selected based on its exceptional performance in achieving a balance between accuracy and speed which is crucial for real-time applications. To optimise the performance of our detection model, we incorporate transfer learning from the classification model. By leveraging the knowledge gained during the classification task, we enhance the accuracy and reliability of our detection system. This approach allows us to capitalise on the strengths of both models and achieve superior results in traffic sign detection and classification.

Keywords: computer vision; traffic sign; YOLOv5.

DOI: 10.1504/IJPT.2024.142180

International Journal of Powertrains, 2024 Vol.13 No.3, pp.269 - 286

Received: 01 Aug 2023
Accepted: 15 Apr 2024
Published online: 11 Oct 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article