A dual-stage deep learning framework for simultaneous fire and firearm detection in smart surveillance systems

Traditional video surveillance systems often treat fire detection and firearm recognition as separate tasks, missing the opportunity to address multiple security threats in an integrated manner. This paper presents a novel dual-stage deep learning framework for real-time, unified detection of fire and firearms in smart surveillance environments. At its core, a Unified Threat Classification Network (UTCN) dynamically routes frames to lightweight YOLOv5n-based fire and firearm detectors, with a Multi-Frame Confidence Evaluator (MFCE) verifying detection consistency to reduce false positives and false negatives. Beyond extensive benchmark testing, the system was validated on unconstrained real CCTV footage and enhanced with a proof-of-concept domain adaptation loop that fine-tunes detection modules using few-shot real-world samples. The proposed framework achieved up to 97.1 % detection accuracy and demonstrated improved robustness in real deployment scenarios, confirming its suitability for scalable, edge-deployable smart surveillance systems.

Abstract licence: CC BY-NC

68 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.