Improved Traffic Sign Detection in Autonomous Driving Using A Simulation-Based Deep Learning Approach Under Adverse Conditions

Autonomous driving faces a significant challenge in recognizing traffic signs in adverse weather conditions. This research aims to address this challenge while also reducing the cost of on-road testing and validation. Current solutions for detecting traffic signs under various driving and weather conditions are deemed ineffective and costly. To overcome these limitations, the proposed method leverages six different weather scenarios to utilize real-time sensory inputs. By integrating YoLov5 vision with a virtual simulation setup, the approach can analyze these inputs and identify traffic signs in a simulated environment using real-captured videos. The model undergoes rigorous testing and validation in a controlled environment, employing a road model and vehicle model to simulate the six different weather conditions. The results demonstrate the success of the approach across all testing conditions, with YoLov5 vision exhibiting a significantly improved level of accuracy in detecting traffic signs, showing an average improvement of 5.56% for day, 7.32% for foggy day, 5.88% for cloudy, 4.55% for dusk, 6.25% for foggy night, and 2.17% for night conditions. Additionally, the research findings highlight the cost-effectiveness of the proposed method compared to existing solutions, emphasizing its potential for enhancing the safety and efficiency of autonomous driving systems.