University AI Surveillance System - Distributed Edge AI for Real-Time Computer Vision
Project Overview
The Challenge
Modern surveillance requires moving beyond passive video recording to active, real-time threat detection. However, streaming dozens of high-definition camera feeds to a centralized cloud for AI processing causes massive network congestion, high latency, and exorbitant bandwidth costs. The objective of this research lab was to design a distributed, Edge AI surveillance architecture capable of ingesting video feeds and executing computer vision models locally.
The Approach
This project was an exploration into high-performance computing (HPC) and localized data ingestion.
Performance Outcomes
Results & Impact
Outcome
Privacy & Ethical Safeguards: Deploying computer vision in a university environment requires strict data governance. A core focus of the architecture was ensuring privacy by design. The system was structured to process and immediately discard raw video frames at the edge. Instead of storing permanent facial biometrics or video archives, the database only logs time-stamped metadata and anomaly alerts, maintaining strict compliance with modern privacy standards.
"The true bottleneck in AI video analytics is rarely the machine learning algorithm itself; it is the hardware infrastructure and the network ingestion pipeline. Optimizing the performance-per-watt ratio on edge computing platforms is the most critical engineering challenge when deploying computer vision systems at scale."