Smart Surveillance AI

The short
version.

A research project on real-time threat detection in public areas — and the false-positive trade-offs that decide whether such a system can be trusted.

Public-space surveillance is already everywhere. The harder question is whether it can be made genuinely useful: can a system identify suspicious activity early, without turning every shadow into an alarm?

This applied research project explores a threat-detection pipeline for public areas, covering model architecture, the labelling strategy behind the training data, and the false-positive trade-offs that separate a deployable system from a demo. The work is documented in a technical report published through Navo Cloud Solutions.

My interest here is not surveillance for its own sake. It is the engineering of judgement under uncertainty — how a model communicates confidence, when a human should be brought in, and how to design a system whose mistakes are visible rather than hidden.

Where the
work lived.

• Computer vision
• Threat detection model architecture
• Dataset labelling strategy
• False-positive analysis
• Responsible deployment

Tools that
earned it.

What I
kept.

A detection system is only as useful as its false alarms are manageable. Trust is an engineering requirement.