Abstract

One of the main goals of the automotive industry is to develop a fully autonomous vehicle. Among the tasks which the fully autonomous vehicle should perform is traffic sign (TS) detection which has a great impact on the behavior of the autonomous vehicle on the road. In this paper, a solution for controlling a vehicle based on detected TSs is presented. For TS detection, the possibility of using state-of-the-art algorithms based on machine learning, specifically YOLOv3 (You Only Look Once) and YOLOv4, was examined. The part of the solution for controlling the vehicle is developed using ROS (Robot Operating System), where the complete solution is tested in the CARLA (Car Learning to Act) open-source simulator. TS detectors are evaluated on a real-world dataset and synthetic dataset, after which the proposed solution is tested on different scenarios created in the CARLA simulator in different weather conditions (sunny, rainy, foggy, night). The results show that the proposed solution achieved the highest correct actions rate of 95.83% when using the TS detector based on YOLOv4 trained on both real and synthetic data.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call