Vehicle Counting System Using YOLOv5 and DeepSORT Algorithms

Abstract

Air pollution is a serious issue in big cities, such as Bandar Lampung. This is caused by high transportation activities using motorized vehicles. Data from 2021 shows a 4.30% increase in the number of motorized vehicles in Indonesia, impacting carbon emissions. In response to this problem, Greenmetric Lampung University has a green transportation work program that focuses on reducing motor vehicle emissions. To support this goal, automatic traffic monitoring is carried out by applying the field of Computer Vision, namely object tracking. The making of an object tracking system in monitoring traffic uses two combinations of the YOLOv5 and DeepSORT algorithms. The method used in this research is the Scrum method which is carried out in three sprints and divided into three stages, namely pre-game, game, and post-game. The result of this research is an object tracking system that successfully distinguishes and counts three types of vehicles (motorcycle, car, and bus) automatically, and has been tested in real-time with an average precision value of 99%, recall of 97%, F1 score of 97.2%, accuracy of 96.8%, and average accuracy of system calculation of 97.65%.