| Title |
Fire and Smoke Detection on an Embedded System using YOLOv8-Nano with CBAM on Hailo-8 |
| Authors |
황호성(Ho Seong Hwang) ; 현예지(Yeji Hyun) ; 김동현(Donghyun Kim) ; 김호철(Ho Chul Kim) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.7.105 |
| Keywords |
Fire detection; Embedded device; YOLO; CBAM; Hailo-8 |
| Abstract |
In this study, the object detection network based on YOLOv8-nano with an attention module was designed and applied for fire detection in an embedded environment equipped with the Hailo-8 accelerator. For the fire detection network, the AI-Hub fire monitoring dataset was utilized, and appropriate preprocessing was performed to enable effective fire detection.Various YOLO architectures were evaluated to compare performance, and YOLOv8-nano was ultimately selected considering accuracy (mAP@0.5)and embedded deployment constraints. To address the critical issue of missed detections(FN) in fire detection scenarios, additional network design modifications were introduced.As a result, the proposed YOLOv8-nano-based network with the CBAM module (YOLOv8n+CBAM) maintains comparable performance to existing models while reducing missed detections to approximately 62% of the baseline level. Furthermore, when the trained network was deployed on the Hailo-8 accelerator, it achieved an average FPS of 144.44, demonstrating its capability for real-time inference. |