Image Recognition for safety on the factory floor
summary
ResNet50 backbone plus a custom U‑style classifier trains on a curated set from >30,000 images sampled every 10s and initially reaches 85% versus a 55% baseline but fails on the empty conel edge case; retraining with ~5,000 re‑sampled images raises accuracy to 96%. Deployment runs on‑prem on Tenova Edge with Azure IoT Edge and a Flask API ingesting an IP camera stream (~300 ms latency), later replaced HTTP with RabbitMQ (Pika) and a pythonnet wrapper for robustness, and triggers an HMI human‑in‑loop stop to avoid $1,000s/min production and electrode damage.
AI-written summary · about
chapters (9)
- 0:00Welcome
- 1:51Tenova overview and motivation
- 4:44Training the model
- 7:56Model deployment challenges
- 12:16RabbitMQ for messaging
- 17:16Python.NET integration
- 19:07Live demo and human-in-the-loop
- 20:32Key takeaways
- 21:36Q&A
related talks
41mPractical Aspects of Applying Modern Computer Vision Technologies
36mAI-Powered Real-Time Quality Control
46mImage Recognition with Deep Learning From prototype to production
14mComputer vision on edge devices
01h 02mAI on the Edge
08mDeep Transfer Learning for Computer Vision
01h 04mBuilding and Training your own Custom Image Recognition AI
40mMachines are Learning