Tata Elxsi's IRIS cuts safety detection to under 5 seconds

Industrial facilities have invested heavily in camera infrastructure over the past decade, yet most footage is recorded and rarely acted on in real time. Tata Elxsi designed IRIS to address that gap by analyzing video as it is produced, without streaming raw video to the cloud. The platform runs in the Asia Pacific (Mumbai) AWS Region, chosen for data-residency requirements and low-latency proximity to customer facilities in India. This setup matters for organizations that operate cameras across multiple sites and need to keep data within specific geographies.

The architecture separates the image path from the metadata path: extracted frames are written to Amazon S3, while the streaming event carries only the S3 object key and context such as camera ID, plant, zone, and an NTP-synchronized timestamp, keeping each event under 1 KB. This design keeps the streaming layer lightweight while models retain full access to the visual data. It may appeal to IT teams looking to scale camera coverage without proportionally increasing cloud data costs.

The platform's correlation layer is what turns raw detections into high-confidence events, reducing spurious alerts by an estimated 40–50 percent compared with passing detections through directly. The outcome hinges on whether this correlation logic holds as sites, lighting, and camera angles vary over time. For plant safety managers and control-room operators, the value is in receiving fewer false alarms that erode trust; for enterprise IT teams, the test is whether the edge-filtering approach maintains detection accuracy while controlling data movement costs.