2 domains
A firearm model and a drone model, trained separately — each with its own dataset, classes and context.
Flagship project · Video intelligence
Cerberus.
The three-headed guardian: detect a firearm or a drone on a video stream, and raise a reliable alert in real time. Computer vision, serving physical security.
YOLOReal-time detection
Temporal alert
The alert only fires after several consecutive detections, to limit false positives on a single frame.
Open & responsible
A detection aid: every alert is verified by a human. Never a single line of defence.
The pipeline
From video to alert.
01. Extract
Frames are extracted from source videos, one every N frames.
02. Annotate
Objects are labelled (CVAT, Label Studio, Roboflow, LabelImg), with background images to reduce false positives.
03. Train
One YOLOv8 model per domain, over several dozen epochs.
04. Detect
Inference on an image, a folder or a video, with temporal-confirmation alerting.
⚠️ Responsible use
Cerberus is an experimental detection aid. It replaces neither a certified alarm system nor human intervention. A model trained on a small dataset produces false positives and false negatives. Any deployment must comply with regulations on video surveillance and data protection (informing filmed individuals, retention periods, etc.).
A custom detection project?
Let's discuss your context: existing cameras, scenes to monitor, field constraints.