International Journal of Computer Science Trends and Technology (IJCST) – Volume 6 Issue 2, Mar - Apr 2018 ISSN: 2347-8578 www.ijcstjournal.org Page 137 Traffic Sign Recognition and classification using Convolutional Neural Networks Kenzari Zouleykha [1] , Boulekouirat Sabrina [2] , Abbas Fayçal [3] , Bekhouche Abdelali [4] Department of Computer Science University of Abbes Laghrour khenchela Algeria ABSTRACT In the field of computer vision the problem of recognition and classification is imposed, the Convolutional Neural Networks (CNN) are a new concept for the deep learning way especially recognition and classification of images and videos, natural language processing and other applications. In this paper we focus on the recognition and classification of traffic sign, Our model operates in two stages, A data processing step to simplify the feature extraction step, the second step is to apply a Convolutional Neural Network to recognize and classify the traffic signs, our method has achieved a low error rate and reached an accuracy of 99.6%. Keywords :— Deep learning, convolutional neural networks, recognition and classification of Traffic Signs. I. INTRODUCTION The recognition of traffic signs is a very important operation for the help and safety of drivers, especially for autonomous cars, the role of a human being would be reduced to simply providing the travel endpoints and some additional requirements, It is very difficult to recognize traffic sign boards because of a lack of visibility such as illumination, objects overlap as well as difficult meteorological conditions. Our approach focuses on the use of the technique of convolutional neural network noted CNN, which is currently the best performing model for object recognition and classification in the computer vision domain. We propose in this work a new approach based on the recognition and classification of traffic signs, it will allow us to identify traffic signs in different meteorological conditions. Our model has proved its efficiency because it reaches a higher precision with a low error rate. The rest of the research is organized as follows: The first section will discuss related work on the recognition and classification of traffic signs. The description of our model will be presented in the second section. In the third section we will present the results obtained and finally we will conclude this work with a conclusion and we will propose some perspectives for future works. II. PREVIOUS WORKS Convolutional networks were introduced for the first time by Fukushima [1] and derived a hierarchical nervous network architecture inspired by Hubel's research [2]. Lecun [3] proposed generalized architectures for classifying digits successfully and for recognizing LeNet-5 handwritten control numbers. Ciresan [4] used convolutional networks and performed best in the literature for multi-object recognition of multiple image databases: MNIST, NORB, HWDB1.0, CIFAR10 and the dataset IMAGEnet. due to the application of the CNN, new methods have been developed by Qian, R. et al [5], To perform an MPPS (max pooling position) recognition as a tool to predict class classifications, MPPS explains the properties between layers based on (GTSRB) database, and improves classification and speed, this has led to an increase in accuracy. Jorge Enrique Zafra et al [6] presents two neural network algorithms for recognition of traffic signs boards using Backpropagation and convolutional neural networks. In training times the Backpropagation network is much faster compared to convolutional neural networks but the accuracy is very high which is equal to 80% and the random precision of Backpropagation which is between 50% and 98%. Lim K, et al [7] presents a real-time traffic sign recognition method based on a graphics processing unit applied to a dataset from Germany and South Korea, They proposed a powerful method against changes of illumination, and they subsequently performed region recognition using a hierarchical model. This method has produced stable results in low illumination environments. The hierarchy of the model was carried out in real time, the proposed method obtained a score of 0, 97 accuracy. A study was presented by Qian R et al [8] to compare four methods, these methods are divided into two categories; the first category is constituted of two descriptors, HP and HOG, the second category is constituted of two classifiers MLP and SVM. The study concluded that the HP- SVM method offers competitive performance in terms of the accuracy and processing time of traffic signs recognition. A recognition system was proposed by Reinders C et al [9] where the advantages of convolutional neural networks and random forests were combined to construct a fully convolutive network for the prediction of inclusive boxes. III. OUR MODEL RESEARCH ARTICLE OPEN ACCESS