Tata Elxsi's IRIS: edge filtering cuts cloud frames 70-80% for real-time factory safety AI

AWS ML Blog · rss · 2026-09-22

AWS ML Blog details Tata Elxsi's IRIS industrial safety platform on AWS. Edge servers (Greengrass + Jetson AGX Orin) extract 2-5 fps frames and filter 70-80% before upload; only <1KB metadata events flow through Kinesis (2,000-5,000 eps, p95 <200ms). SageMaker endpoints per model family (YOLOv8 PPE detection, polygon geofencing, SlowFast action recognition, monocular-depth vehicle proximity) run on ml.g5.xlarge at 40-60 RPS with p95 <300ms, scaled by target tracking plus SQS queue-depth policies and 4-8 frame micro-batches. A fully serverless, event-driven blueprint decoupling image and metadata paths.

Original post →

More from Infra

Infra channel →