New Deep Learning Architecture for Improving the Accuracy and the Inference Time of Traffic Signs Classification in Intelligent Vehicles

Vehicular Ad-hoc Network (VANET) is a new technology on which are based Intelligent Transportation Systems (ITS). The goal of this technology is to improve the vehicular environment, and to provide more safety for both vehicles and drivers. In this global context, characterized by the use of IoT, vehicles are integrating more intelligent tools that help to insure vehicle-to-vehicle, vehicle-to-infrastructure and vehicle-to-driver communications. Hence, IoT involves the necessity to handle and manage a huge amount of data, including processing and analyzing traffic signs and road scene images. In fact, dealing with these aspects of Big Data (volume, variety, etc.) presents a big challenge, for which many approaches are proposed, including Deep Learning (DL). Although their high performances, this type of approaches still faces many difficulties, which are related essentially to the computational load and the hardware requirements. From this perspective, we have adopted a new Deep architecture to ensure traffic signs classification. The objective of the proposed architecture is to speed up the training and the inference stages, and that without affecting classification’s performances. The obtained results show that the adopted approach accelerates the training & the inference speed, and reaches high accuracies using a limited number of parameters. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.