Adaptive Traffic Signal Control System Using Object Detection Based Approach

In Bangladesh, most four-legged intersections use a static-timed or manually-controlled traffic signaling approach, which creates irrefutable congestion and requires human involvement. This paper proposes an adaptive traffic control system utilizing the YOLO object detection framework within CNNs to address this issue. The system captures and frames images using cameras facing every road at a roundabout to detect and count vehicles in real-time for providing lights based on the traffic conditions, prioritizing emergency vehicles such as ambulances, firetrucks, and police-car while also maintaining queues. It sets the signal duration considering road condition. Using a dataset of 10,462 images across 12 classifications of automobiles and a transfer learning approach in YOLOv8, the model achieved a mAP50 score of 0.90. This automated method reasonably enhances traffic flow management by dynamically adjusting signal lights without human intervention.