Safety and Occupancy with AI
summary
A custom vision model was trained in Vertex AI/AutoML on labeled door-entry clips (≥100 examples per label), compiled to an Edge TPU TensorFlow Lite blob and deployed to Coral hardware at the camera to run inference on-device, emitting only numeric events (no video) every ~15 seconds. Event summaries are sent through IoT Core → Pub/Sub → Cloud Functions into BigQuery and Looker for occupancy and safety analytics; internal tests detected entry/exit and a demo reported one employee and two customers.
AI-written summary · about
chapters (8)
- 0:00Edge AI for occupancy tracking
- 2:18Building AI with Vertex AI AutoML
- 5:40Testing on Coral Edge TPU
- 7:08Streaming inference to Cloud
- 9:43Storing data in BigQuery
- 10:50Merging AI data with BlueZoo sensors
- 12:23Analyzing occupancy trends
- 13:16Drilling into store areas
related talks
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
47mProduction ready EdgeAI with Eclipse ioFog and Google Coral
55mEdgeAI - AI in the Palm of Your Hand
14mComputer vision on edge devices
47mUnlocking Data-Driven Apps with TensorFlow
36mAI-Powered Real-Time Quality Control
06mAI in production: The droids you’re looking for
43mIntersection of Awesome: Building AI enabled IoT solutions powered by the cloud