Traffic sign recognition using deep learning

Abstract

Traffic sign recognition plays a significant role in modern automated driver assisting systems and showing information about safety measures. It is a technology that allows users to recognize traffic signs in real-time, typically in videos, or sometimes just in photos. Poor identification of traffic signs cause road accidents. Moreover In adverse situation like heavy rain,foggy weather or sleepy driver can misidentify a traffic sign that may cause the death of hundreds of people. As a result identification of traffic signs properly has become an obligatory topic for research. In this research, we have used convolutional neural network for detecting and classifying the road signs accurately. We have proposed five Keras models of CNN and compared their results. The main challenge of this research is dealing with noise in images such as ads, parked vehicles, pedestrians, and other moving objects or background objects that made the recognition much more difficult. Not only the objects but also various environmental issues like the reflection of light, rainfall, fog etc has affected the research. In order to conduct this research we have collected our own data-set. We roamed around Dhaka city and clicked pictures of the traffic signs as there is no benchmark data-set available in the perspective of Bangladesh. For 500 images this model gives out an accuracy of 63%. There have been many researches in this field but our one is unique as it is tested on our own collected data-set on Bangladesh’s perspective. Recognizing traffic signs has become a part of our daily essentials as road safety depends on it, on a large scale which made it an obligatory topic for research.