#31 - FridgeWatch

Surveillance Dashboard
Incident Analysis
Incident Report
Heatmap
Evidence Explorer
System Log

Inspiration

Someone walked up to our table and stole our Doritos in broad daylight. We had no proof, no recourse, and absolutely no chill about it. So we did what any reasonable engineering team would do—we built a military-grade forensic surveillance platform to make sure it never happens again.

What It Does

OverSight turns raw surveillance footage into structured, explainable, court-admissible intelligence. It tracks individuals across multiple camera feeds, builds millisecond-precise forensic timelines, and allows investigators to query evidence in plain English.

The same system that catches a Dorito thief scales directly to critical infrastructure breaches, government security, and defense investigations. The underlying problem is always the same: someone was somewhere, something happened—prove it.

How We Built It

OpenCV forms the backbone of our real-time video processing pipeline, handling frame extraction, motion detection, and core computer vision functionality.
React powers the investigation dashboard, including the forensic timeline, bounding box overlays, and AI-powered query interface.
TensorFlow drives movement detection and anomaly flagging by learning normal scene behavior and surfacing deviations in real time.
Python handles heatmap generation by mapping trajectory coordinates onto a 2D floor plan using homography transformations.

Accomplishments

We are most proud of building our own detection algorithm from scratch—not wrapping an existing API or fine-tuning a pretrained black box, but writing the core logic ourselves and watching it work on live footage under real hackathon conditions.