Deep Learning-Based Road Traffic Density Analysis and Monitoring Using Semantic Segmentation

Abstract

Due to factors such as a growing population, more people using private vehicles, and outdated transportation infrastructure, Jakarta, the capital city of Indonesia, suffers from chronic traffic congestion. The environment, citizens' safety, productivity, and quality of life are all negatively impacted by these interruptions. In response to these difficulties, this study proposes a novel method for traffic monitoring. By combining YOLOv5, optical flow, and recurrent neural networks (RNN) with image processing and artificial neural networks, a unified traffic monitoring system can be achieved. We went with YOLOv5 because of how well it identifies various automobiles. The number of vehicles is counted between video frames using Optical Flow, and then the traffic density is classified using RNN. With an accuracy of 87% following testing, RNN was clearly a winner when it came to vehicle density classification. The goals of this research are to lessen the societal and environmental toll of traffic congestion, increase our knowledge of and ability to control Jakarta's traffic, and lay the groundwork for the creation of more advanced traffic monitoring systems. The growing traffic issues in the nation's capital are anticipated to be alleviated with this strategy.