Industrial PPE Compliance AI: Real-Time Computer Vision & Safety Monitoring

Across heavy manufacturing complexes, petrochemical refineries, and high-rise construction sites, strict adherence to Personal Protective Equipment regulations forms the critical baseline for preventing catastrophic injuries and regulatory penalties. However, conventional manual safety inspections suffer from structural vulnerabilities that leave facilities exposed to severe operational hazards.

First, manual safety audits offer minimal temporal and spatial coverage, typically inspecting less than five percent of total working hours across vast physical layouts. Second, human safety officers inevitably face perceptual fatigue in busy environments, frequently missing critical transient violations such as discarded helmets in elevated zones or missing reflective vests amidst dynamic vehicle traffic. Third, traditional paper-based incident logging lacks objective visual verification, creating retrospective disputes during compliance reviews rather than driving preventative accountability.

To eliminate these blind spots, I designed and developed Industrial PPE Compliance AI as an autonomous, edge-capable vision system that connects directly to industrial CCTV cameras and network streaming protocols:

The ingestion layer implements a dedicated, non-blocking video capture thread protected by mutual exclusion locks. This architecture prevents network buffer accumulation common in RTSP streams, ensuring that the computer vision pipeline always evaluates the freshest optical frame without latency accumulation over extended runtimes.

Captured frames pass into a multi-class YOLO neural network fine-tuned to classify human figures alongside individual safety gear components, including hard hats, safety vests, protective goggles, and industrial boots. To preserve temporal identity across noisy factory floors, the system couples object detections with a persistent multi-object tracking engine, maintaining worker identities and movement vectors across consecutive frames.

Rather than relying on naive proximity assumptions, the platform executes a sophisticated geometric containment pipeline to verify that safety equipment is correctly fitted:

The platform provides comprehensive situational awareness for safety superintendents, site engineers, and system integrators:

The web-based HSE operations dashboard features a modern glassmorphism design streaming live video via multipart MJPEG with dynamic bounding box overlays. Safety supervisors can monitor live compliance percentages, inspect hourly violation distributions, and resolve open incident flags with one-click audit actions.

For retrospective investigations and automated compliance reporting, the system includes a robust command line tool built with Click and Rich for batch analyzing recorded surveillance footage, alongside fully documented FastAPI REST endpoints containerized with Docker for rapid industrial edge deployments.