Deep learning system detects smoking in fire exits using YOLO models

A research paper details a deep learning system designed for real-time smoking detection in fire exit zones using CCTV surveillance. The study evaluated YOLOv8, YOLOv11, and YOLOv12, developing a custom YOLOv8-derived model that achieved a recall of 78.90% and mAP of 83.70%. Performance tests on edge devices indicated the system is suitable for time-sensitive operations, with the Jetson Xavier NX processing data between 52 to 97 milliseconds per inference. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enhances public safety monitoring capabilities by providing automated detection of fire hazards in critical areas.

RANK_REASON This is a research paper detailing a novel application of deep learning for safety monitoring.

Read on arXiv cs.CV →

• YOLOv8
• YOLOv11
• YOLOv12
• Jetson Xavier NX
• arXiv

• paper
• safety