On-Site Video Analytics on ATLAS BOX

The problem

Analysing video centrally means moving it there first. On a site with limited connectivity that is either expensive or impossible: continuous streams from several cameras do not fit a thin uplink, and when the link drops the analytics stop with it. Closed and isolated networks rule the approach out entirely. Recording locally while analysing centrally splits the system in two, leaving the footage in one place and the detections in another.

The solution

Video analytics and video management both run on the ATLAS BOX appliance, next to the cameras.

• Runs detection and tracking on the appliance, so only results leave the site — if anything does.
• Records and plays back video on the same appliance, keeping footage and detections in one system.
• Holds recorded video on 4TB of local NVMe storage.
• Processes inference on a 6+ TOPS NPU, with an ARM A76 CPU running the applications.
• Operates fully offline, including on an air-gapped network.
• Runs face recognition alongside the analytics on the same box, where a site needs both.
• Keeps analysing through a connectivity outage, because the uplink is not in the processing path.

What we brought

• Detection and tracking
• Video management
• Local recording
• Offline operation

How we solved it

1. 1Capture
2. 2Analyse
3. 3Record
4. 4Review
5. 5Act