SmartSite Vision — Edge Computer Vision & Safety AI

Industrial job sites operating multi-shift workforces face significant occupational safety and OSHA compliance liabilities. Despite strict safety protocols, manual supervision is inherently intermittent: site leads can only observe one zone at a time, and violations during high-activity periods frequently go unnoticed until an incident occurs. Compounding safety insurance rates and regulatory scrutiny require continuous, automated detection that catches hazardous zone breaches and gear violations in real time rather than logging them reactively after an audit.

We engineered SmartSite Vision as a distributed edge-AI safety platform. Custom YOLOv8 object detection models were fine-tuned on a curated dataset of 48,000 labeled construction site images covering 11 PPE item classes — hard hats, safety vests, steel-toed boots, gloves, eye protection, and fall harnesses. The models run natively on edge compute nodes mounted behind IP cameras, eliminating cloud-inference latency entirely. DeepSORT tracking assigns persistent IDs to workers across frames so the system monitors cumulative exposure in hazardous zones — not just point-in-time violations. When a violation is detected, a FastAPI event pipeline triggers a sub-400ms Slack/SMS alert to the site supervisor with a timestamped annotated image. The Next.js command dashboard provides live multi-feed monitoring, historical incident heatmaps, worker compliance scores, and automated OSHA-format incident logs.