Detection and Recognition Of Traffic Sign By Convolution Neural Network

Abstract

Modern automated driver assistance systems that display safety information heavily rely on Traffic Sign Recognition and Detection. It is a system that enables users to instantly identify traffic signs, usually in films but occasionally only in still images. Road accidents are caused by improper interpretation of traffic signs. Moreover hundreds of people could be killed if a driver misidentifies a traffic sign in hazardous conditions like heavy rain, cloudy weather, or sleepiness. As a result, the appropriate identification of traffic signs has become a required research issue. Convolutional neural networks were employed in this study to accurately detect and classify the traffic signs. Five Keras models of CNN have been proposed, and their output has been compared. Dealing with picture noise, such as ads, parked cars, pedestrians, and other moving things or background objects that make recognition considerably more challenging, is the key difficulty of this research. The investigation has been impacted not only by the objects but also by a number of environmental factors as light reflection, precipitation, fog, etc. We have assembled our own data-set in order to undertake this research. We wandered the streets of Chittagong and took images of the traffic signs because there is no benchmark data set for Bangladesh accessible. This model provides a 98% accuracy for 39,200 photos. Numerous studies have been conducted in this area, but ours stands out since it is based on data that we have independently acquired from Bangladesh.