Danger Detection with ML
Real-time computer vision prototype that detects firearms and fire in video streams and sends alerts with visual evidence
Danger Detection is an open-source prototype that detects guns, rifles, and fire in video streams and sends an email alert with a captured image. I first built it for a competition in 2021 and later refactored and documented it for public use.
It received a GW Open-Source Student Award in 2026, and I presented it at the 2026 GW Open Source Conference in the lightning talk “Open Source AI for Real-Time Danger Detection.”
How It Works
A YOLOv3 model, run through OpenCV, processes frames from a webcam or video stream. When it detects a threat, the system captures the frame and sends a notification by email. It averages 25 frames per second on a CPU, and GPU acceleration is available through CUDA.
Built With: Python, OpenCV, YOLOv3, NumPy, SMTP