Learn how our hybrid edge architecture turns standard IP cameras into autonomous hazard detection sentinels without streaming sensitive raw video over public networks.
Your on-site IP cameras broadcast standard RTSP/ONVIF streams directly into our ruggedized edge processor at 30 FPS.
YOLOv9 models execute on NVIDIA TensorRT hardware in 35ms, performing multi-class bounding box regression for gear and equipment.
Perspective geometry translates 2D pixel feeds into 3D CAD coordinates, checking against rotating crane radiuses and open floor voids.
When an anomaly violates safety rules, the system dispatches push notifications, SMS alerts, and Procore tickets in under 200ms.
Standard off-the-shelf vision models fail on construction sites due to optical reflections, heavy dust, workers carrying drywall, and shadows. RigSafeLens utilizes temporal tracking and site-specific color harmonization to virtually eliminate false alarms.
{
"timestamp": "2026-10-21T12:44:02.194Z",
"camera_id": "CAM-NORTH-TOWER-L18",
"inference_latency_ms": 38.4,
"detections": [
{
"worker_id": "WKR-8812",
"hard_hat_present": false,
"confidence": 0.9942,
"high_vis_vest": true,
"elevation_hazard_proximity_m": 0.82
}
],
"safety_evaluation": {
"severity": "CRITICAL_HAZARD",
"osha_code_violated": "29 CFR 1926.502",
"action_triggered": "SUPERVISOR_TABLET_DISPATCH"
}
}