Comparison of ResNet & VGGNet for Traffic Sign Image Quality in Nighttime

Traffic signs are one of the important road equipment facilities that inform road users about regulations and instructions visually. Currently, an automated Traffic Sign Recognition (TSR) system is being developed and applied to advanced driver systems (ADAS) so that road users can be safe and secure while on the road. This study explores the application of Histogram Equalization (HE) to improve the visibility of traffic signs in low-lighting conditions, using the deep learning architectures ResNet50 and VGG19. This study aims to evaluate the impact of HE on the performance of these models in recognizing traffic signs. The experimental results show that on the ResNet50 architecture, HE maintains the test accuracy at around 92%, while on VGG19, HE improves the test accuracy from 89.62% to 92.54%. These findings are relevant for Traffic Sign Recognition (TSR) systems, which are an important component of Advanced Driver Assistance Systems (ADAS). By improving traffic sign recognition accuracy under low illumination conditions, Histogram Equalization contributes to improved road safety through more reliable and responsive ADAS systems, which rely on TSR to provide critical information to drivers.