How Tata Elxsi detects industrial safety risks in seconds on AWS

This article describes how Tata Elxsi built IRIS, a real-time industrial safety platform on AWS that detects unsafe conditions in seconds rather than minutes using computer vision and edge processing.

• Edge servers extract frames and filter with motion detection, uploading only safety-relevant frames to S3
• Metadata streams through Amazon Kinesis Data Streams at 2,000–5,000 events per second
• Custom computer vision models on Amazon SageMaker AI detect PPE compliance, restricted-zone intrusions, unsafe postures, and equipment proximity
• Correlation layer in Lambda and DynamoDB reduces false positives by 40–50 percent using temporal patterns and zone context
• Alerts escalate by severity through EventBridge and Step Functions to dashboards, email, SMS, and mobile notifications
• All events stored in S3 with lifecycle policies for compliance and audit trails
• Production results: detection under 5 seconds, 15–20% reduction in safety incidents, 30% lower surveillance costs

IRIS demonstrates how existing camera infrastructure becomes a proactive safety system by filtering at the edge, streaming metadata instead of video, and correlating detections into high-confidence events.