An Internet of Things-based Traffic Light Control System Utilizing Computer Vision for Real-time Traffic Density Calculation

Abstract

The use of traditional traffic light timers in the Philippines encounters limitations in dealing with dynamic fluctuations of traffic density. This observation led to the study of innovating a traffic light to be adaptive. While most existing research has focused on vehicle counts as a variable to adjust the traffic light based on traffic condition, limited studies have incorporated utilizing computer vision to consider vehicle occupancy for the same purpose. Additionally, the use of emerging computer vision technology in this respect recommends further exploration. In response, the researchers proposed this study that presents a system that can adjust the allocation of traffic light timer based on the calculated traffic density via computer vision. The system gathers real-time traffic feed with utilization of a YOLOv8 based model to detect and classify vehicles and identify the traffic conditions. A prototype is developed to simulate the traffic light connected via Internet of Things (IoT). Results showed that the model can detect and classify vehicles into four classes with 0.94 mAP@0.50. This presents the potential of the system to be a precursor for the actual implementation of adaptive traffic lights and improve traffic management.