Indonesian Traffic Signs Recognition Using Convolutional Neural Network

Abstract

A traffic violation is one of the reasons for the increasing mortality every year. Traffic Sign Recognition (TSR) is an important component within the scope of Advanced Driver Assistance System (ADAS) and autonomous vehicles, which concern the problem of Traffic Sign Classification (TSC) and Traffic Sign Detection (TSD). The detection system of traffic signs purpose is to warn the driver about the traffic condition they will pass, so it can help the driver to decrease the accident. The traffic signs in every country have different shapes and colors, so the traffic signs have wide variability. This paper shows a traffic signs classification study using a Convolutional Neural Network (CNN). The data is from the traffic signs in Indonesia which consist of 41 traffic signs. The model proposed has shown good enough performance in which the accuracy score is 93% and the average F1-score is 94% in the recognition of traffic signs in Indonesia.