Computer Vision / Applied AI
Smart Surveillance System
Multi-camera detection and tracking with YOLOv8, built as an applied computer-vision project.
Problem
An exploration of real-time, multi-camera video analysis: detecting and tracking people across camera feeds, logging entry/exit events, and generating alerts on defined conditions — the kind of pipeline that sits behind physical-security tooling.
Architecture
Detection (YOLOv8) and tracking are separated from the event/alerting logic, so the alerting rules can change independently of the underlying detection model.
Hover or tab through a node to see what it does.
What it does
- IMPLEMENTED: real-time object detection with YOLOv8
- IMPLEMENTED: multi-camera video processing
- IMPLEMENTED: entry/exit event logging to MongoDB
- IMPLEMENTED: alert generation on defined trigger conditions
Engineering decisions
YOLOv8 for detection
A well-supported real-time detection model that gives a good accuracy/latency tradeoff for multi-camera processing without needing custom model training as a starting point.
Separate the event-logic layer from the detection layer
Keeps alerting rules (what counts as a notable event) swappable without touching the detection/tracking pipeline.
Scope note
FUTURE / RESEARCH SCOPE (not implemented in the current build): face-recognition-based identity matching, criminal-face-database matching, and edge-deployment optimization are directions for this project, not shipped capabilities. They are listed here explicitly so the distinction between implemented and future work is clear. This system has not been deployed in any real-world security context.