CCTV Smoke & Fire Emergency Detection Dataset

The most critical fires are the ones detected in their first 60 seconds. While most fire datasets focus on large-scale forest fires or fully developed structural conflagrations, this dataset is specifically designed to train models to identify early-stage ignition: a smoking trash bin, a small paper fire on a sidewalk, or a discarded cigarette smoldering in the grass — all captured from the realistic, high-angle perspective of a Public Video Surveillance (CCTV) camera. Rendered from 10–20 ft mounting heights with realistic CCTV distortion, compression artifacts, and low-light noise. Incident categories include bin fires (internal ignitions and smoke rising from public or industrial waste containers), ground ignitions (small-scale paper, cardboard, or debris fires on asphalt and concrete), vegetation smoke (small spot fires or smoldering cigarettes in grass or public park areas), and urban contexts (alleyways, bus stops, plazas, and commercial exterior zones). 100% synthetic — no real public areas, PII, or real-world incidents are depicted, ensuring full GDPR compliance for AI R&D. The open-source sample includes 220 images; the full package contains 2,500 images covering 50+ diverse urban and industrial scenarios.