Traffic Sign Recognition for Autonomous Vehicles Using Convolutional Neural Network

  • Animesh Gupta, Mrs. C. Jayavarthini, Chetan Asrani

Abstract

This paper proposes a traffic sign recognition system for autonomous vehicles using convolutional neural network (CNN). With the growing population and increased traffic on roads, number of accidents have significantly increased over the last decade. Self-driving vehicles seems to be an efficient way to tackle this issue. Traffic signs play a very important role in maintaining smoothness of traffic flow. Shape and color are the 2 major factors that distinguishes traffic signs from rest of the objects.The system that has been proposed in this paper involves 4 stages: 1) Pre-processing stage which uses algorithms like gamma correction, geometric transformation and histogram equalization. 2) Color segmentation and object filtering stage to remove unwanted noises from the images and normalize the dataset. 3) Training stage which uses CNN paired with a linear classifier. Training data needs to be scaled, normalized, extended and augmented. 4) Validation and testing stage in which traffic signs are recognized and accuracy is calculated. From the results we conclude that our system achieved higher accuracy due to architectural optimization of the neural network for traffic sign recognition and due to the different type of techniques that we used for pre-processing of images. The techniques used by us made our system achieve better accuracy under variable lighting conditions. Our proposed system performed significantly better on partially occluded and blurred images. The dataset our system uses for training the model is GTSRB (German traffic sign recognition benchmark) dataset.

Published
2020-06-01
How to Cite
Animesh Gupta, Mrs. C. Jayavarthini, Chetan Asrani. (2020). Traffic Sign Recognition for Autonomous Vehicles Using Convolutional Neural Network. International Journal of Advanced Science and Technology, 29(08), 2180 - 2189. Retrieved from https://sersc.org/journals/index.php/IJAST/article/view/23353
Section
Articles