Conventional fixed-time traffic signal systems are structurally incapable of responding to the stochastic variability of urban traffic flow, resulting in prolonged vehicle waiting times, suboptimal intersection throughput, and unnecessary fuel consumption. This paper presents a complete, edge-deployed adaptive traffic signal management system that uses real-time video input from existing CCTV infrastructure and the YOLOv8 deep learning object detection model to continuously estimate lane-wise vehicle density and dynamically compute optimised signal phase durations. The architecture is modular, comprising video acquisition, frame preprocessing, YOLOv8-based vehicle detection and classification, density estimation, decision logic, and signal control modules. The system avoids cloud dependency through localised edge processing, ensuring end-to-end signal update latency below 250 ms. Experimental evaluation across four simulated intersection lanes demonstrates an overall vehicle detection mAP@0.5 of 92.9% at 46 frames per second, a 38.3% reduction in average vehicle waiting time, and a 37.5% improvement in intersection throughput relative to a fixed-time baseline. Comparative benchmarking against Faster R-CNN, SSD, and YOLOv3-based approaches confirms the superiority of the proposed implementation on both detection accuracy and real-time responsiveness. The system is deployable without additional roadside hardware investment, making it a cost-effective and scalable solution for smart urban traffic management.
Ashwin et al. (Thu,) studied this question.