Project overview
An AI-powered camera system for public safety and security surveillance: multi-threaded YOLO detection over live feeds, with remote rPPG pulse extraction reading heart rate from video alone.
The system leverages Ultralytics YOLOv8, trained on over 125,000 images across multiple datasets. The model detects people, vehicles, drones and weapons including handguns, rifles and grenades. What sets it apart is the ability to process video feeds at low latency while holding accuracy on threat detection.
The work covered the core computer vision algorithms and optimising the models for real-time processing. Alerts are integrated end to end: email notifications and the Twilio API for automated calls, so a detection becomes a phone call rather than another line in a log nobody is reading.
Project gallery
Click any image to enlarge.