Problem or Motivation
Workshop safety compliance checks can be inconsistent and reactive. I wanted to build a lightweight system that helps teams identify risk situations earlier.
What I Built
- Dataset preparation workflow for PPE and non-PPE samples.
- Vision inference service with rule-based alert conditions.
- Simple web dashboard for live feed status and event logs.
Tools and Technologies Used
- PyTorch, OpenCV, Python
- Flask API and lightweight web interface
- Deployment experiments on edge and cloud environments
My Role and Contributions
- Handled model training, inference optimization, and event logic.
- Designed user-facing alert workflow and dashboard view.
- Measured performance tradeoffs between accuracy and latency.
Challenges and Lessons Learned
- Lighting variation significantly affected detection consistency.
- Small datasets required augmentation and careful validation splits.
- Practical safety tools need low false alarms to maintain trust.
Gallery or Screenshots
