TRANSFER LEARNİNG-BASED DEEP NEURAL NETWORK FOR TRAFFİC SİGN CLASSİFİCATİON

Abstract

Traffic signs play a major role in human life while travelling ahead. Traffic signs are represented by specific symbols. These signs guide road conditions, provide instructions, give warnings to the drivers. As self-governing cars also came into form its mandatory for those cars to identify these signs. Now-a-days, the classification of traffic signs is a requisite for both the self-governing cars and driver assistance systems as the traffic signs vary from country to country. It is a field of computer vision. This system could guarantee the life of a human and enhance the road traffic safety situations. It assists the drivers or self-governing driving systems by detecting and classifying the traffic sign. It also alerts the driver or self-driving systems in a correct path by classifying the traffic sign boards. In this paper, Convolutional Neural Networks (CNN) is used for image classification. Here, German Traffic Sign Recognition Benchmark (GTSRB) Dataset is used which is open source. This dataset contains 50000 images as due to our system specifications, only 4300 images were taken and classified them into 43 classes where each class contains 100 images. This dataset is processed over two pre-trained models VGG-19 and VGG-16 along with CNN. The images which are categorized into 43 classes are passed as inputs and the output will show to which traffic sign the image belongs. Among these two pre-trained models VGG-16 gained 95.93% accuracy.