Mustafa Abdulkhader Jassim

Abstract :

The high rate of urbanization of Baghdad City has augmented the level of traffic congestion challenges that require smart real-time monitoring systems that can work within limited computational and energy budgets. The presented paper suggests an Energy-Considerate Edge AI Framework of real-time object detection based on traffic cameras mounted in the road network of Baghdad. The framework combines a lightweight YOLO-based detector with an adaptive Energy-Aware Scheduling Module (EASM) that will dynamically adapt inference frequency based on real-time power telemetry and an estimate of scene complexity. Experimental tests of traffic images at various Baghdad crossing points prove that the proposed framework yields an average Precision (mAP) of 87.3% and less energy consumption (up to 43% less than traditional cloud-offloaded systems). The system maintains an average throughput rate of 28 frames per second on embedded edge hardware, and meets with real-time standards of operation at varying levels of traffic density and lighting load. The contribution of this work is a blueprint of smart city traffic surveillance that is scalable and practically deployable, specifically in the context of the constraints of resources in urban areas of developing countries.

Keywords:

Edge AI, YOLO, Object Detection, Energy Efficiency, Traffic Monitoring, Baghdad, Smart City, Real-Time Inference