Framework for Adaptive Traffic Light System

Smart cities strive to integrate multiple sectors such as traffic, energy, healthcare, and governance to enhance urban living. However, the escalating number of vehicles on roads without an equal increase in road capacity leads to congestion, accidents, environmental problems, and a decline in quality of life. Addressing these challenges requires effective traffic management, notably through dynamic scheduling of traffic lights, to mitigate these concerns and ensure smoother urban mobility. The current manual scheduling of traffic lights, which relies on static timing, is inadequate and unable to cope with unpredictable traffic conditions. This inefficiency results in wasted time, energy, and negative economic implications. Utilizing object detection for traffic monitoring p resents a potential solution to address these challenges. Thus, this paper introduces a framework for detecting and monitoring traffic. To do so, the paper outlines the following contributions:: Firstly, it presents a dataset comprising various classes of vehicles. Secondly, it conducts a comparative analysis among four commonly used computer vision models: YOLOv8, YOLOv9, Faster R-CNN, and Rt-Detr. Thirdly, it introduces a simulation environment aimed at comparing and determining the most effective scheduling technique for traffic lights based on Round-robin scheduling. The results indicate that YOLOv8 with instance segmentation methodology achieves the highest mean Average Precision (mAP) at 94.8%, followed by YOLOv8 with object detection methodology at 88%, and YOLOv9 at 88%. Additionally, the simulation environment is evaluated using synthetic data of different scenarios to assess traffic scheduling. The primary result demonstrates a 33.47% reduction in time compared to static systems.