CNN-Based Traffic Sign Recognition
Shin Wee Fiona Liou, Hau-Lee Tong, Kok-Why Ng, Hu Ng · 2022
Traffic signs are a crucial part of maintaining driver and pedestrian safety on the road since they are being designed to provide essential information and alerts of potential hazards.With the rapid development of Advanced Driver Assistance Systems (ADAS), traffic sign recognition is also becoming much of a concern.However, due to real-world variations such as lighting conditions, occlusion, weather factors, motion blur and colour fading, there are still some failures in traffic sign recognition that cannot be perfectly resolved.Therefore, we implement image enhancement techniques and a pre-trained convolutional neural network for traffic sign recognition in this paper.Our proposed model uses the pre-trained VGG19 model as the baseline model and changes the fully connected layer and classifier of the VGG19 model.The experimental results demonstrate the effectiveness of applying image enhancement.Our proposed model was able to outperform the traditional machine learning method but did not surpass other deep learning methods.