A Robust Resource-Aware Framework for YOLO-Based Object Detection in Dynamic and Multi-Platform Resource-Constrained Environments

Efficient management of resources such as CPU, RAM, and disk storage is a critical challenge for object detection systems that rely on multi-camera video streaming, especially for resource-constrained devices. These constraints frequently result in the risk of overheating and potential system failure during high workloads. This paper presents a novel Resource-Aware Framework (RA-Framework), designed to optimize resource allocation using dynamic adaptation mechanisms. The RA-Framework dynamically manages CPU, RAM, and disk storage resources for YOLO-based object detection, supporting multiple camera streams and scheduling mechanisms. The framework was implemented in resource-constrained environments using various YOLO models with single and dual-camera setups. Experimental results showed that the RA-Framework successfully reduced average CPU usage by up to 68.62%, decreased RAM consumption by approximately 13.47%, and extended the time for disk usage to increase by 1% by more than 500% in specific scenarios. In contrast, systems without the RA-Framework experienced higher resource utilization and greater susceptibility to resource scarcity or even overheating.