AFAR-YOLO: An Adaptive YOLO Object Detection Framework

Abstract

This study focuses on developing an advanced early warning system utilizing YOLOv5 to detect objects indicative of potential fire hazards. This research is motivated by the fact that continuous monitoring is impractical, especially in high-risk and inaccessible areas. We introduce an innovative approach: adaptive YOLO for object detection to enhance early fire detection capabilities in these challenging environments. The main contribution of this research is the development of adaptive frames per second (FPS) resolution in YOLO object detection. We found that implementing adaptive FPS alone does not significantly impact the efficiency of CPU and RAM resources in the tested devices. However, when adaptive FPS is combined with adaptive resolution, resource usage is significantly reduced–specifically, a 33% decrease in CPU usage and a 0.5-1% (200-400 MB) reduction in RAM usage. These efficiency gains are important in enhancing safety in the industrial sector.