The Challenge
Warehouse environments are inherently dangerous. Human safety monitors can’t watch every camera feed simultaneously, and incidents often happen in blind spots or during shift changes. Reactive safety programs only kick in after someone gets hurt.
Approach
We partnered with a warehouse operator to build a proactive vision system — from evaluating the best AI models, through collecting and annotating real warehouse footage, to deploying a live monitoring application.
What We Built
Research and evaluation of the best AI vision models for warehouse hazard detection
Custom dataset of annotated warehouse footage teaching the AI safe vs. unsafe conditions
Fine-tuned model that reliably spots safety issues in real-world conditions
Landing page communicating the product to stakeholders and customers
Live safety monitoring application analyzing camera feeds and flagging hazards in real time
Technologies & Methods
Selected Visuals
