Indonesian Traffic Sign Recognition For Advanced Driver Assistent (ADAS) Using YOLOv4

Abstract

Traffic violations are one of the causes of the increasing number of road traffic fatalities every year, apart from driver negligence or ignorance of traffic signs. ADAS does not totally forestall mishap, however they can all more likely shield us from a few human elements and human mistake. The goal of ADAS is to automate vehicle systems for better driving and safety, such as Traffic Sign Recognition (TSR). This paper presents a study to recognize traffic sign patterns using YOLOv4 using the Indonesia Traffic Signs (ITS) dataset. The ITS dataset consists of four categories (warning, prohibitory, mandatory and directive) with twenty six signs. The deep learning model of YOLOv4 is based CSP-DarkNet53 backbone which has shown good performance with main Average Precision (mAP@0.5) of 74.91% for 26 signs of Indonesian Traffic Signs.