AI-Powered Smart Traffic Monitoring System Using Deep Learning and Cloud Integration

Abstract

Urban traffic congestion presents a critical challenge for modern smart cities as vehicle populations continue to expand exponentially. Conventional fixed-timer traffic signal systems allocate equal green durations to all lanes irrespective of real-time traffic density, producing unnecessary delays, fuel wastage, and elevated emissions. This paper presents an AI-powered Smart Traffic Monitoring System that integrates deep learning-based object detection with cloud-enabled data analytics to enable fully adaptive signal control. The proposed framework employs YOLOv8 to detect and classify vehicles—including cars, buses, trucks, motorcycles, and ambulances— from traffic camera imagery. The intersection frame is partitioned into four quadrants representing independent lanes; vehicle centroids derived from bounding-box coordinates are used to assign counts per lane. An adaptive green-time formula computes proportional signal durations based on lane density, while a dedicated emergencyvehicle module overrides normal scheduling to grant immediate priority when an ambulance is detected. Annotated output images, lane statistics, and recommended signal timings are delivered through a Flask-based web application, and all prediction records are persisted in a JSON data store for cloud-accessible historical analytics. Experimental results demonstrate vehicle-detection accuracy exceeding 91%, mean image-processing latency below 1.8 seconds, and a measurable reduction in average lane waiting time compared to fixed-cycle baselines. The system provides a scalable, cost-effective foundation for intelligent urban traffic management aligned with smart-city objectives