Cameras that see, count, and decide — at the edge, in real time. Six productized verticals on NVIDIA Metropolis: smart city, industrial, healthcare, retail, defence, and agriculture.
Each vertical is a working product, not a custom build. Configurable, deployable, supported.
Crowd density, traffic flow, incident detection, and infrastructure inspection. Integrated with municipal command centers and emergency services.
Defect detection, PPE compliance, and predictive maintenance from visual signals. Built for refinery, petrochemical, and manufacturing floors.
Radiology triage, patient-flow analytics, and ICU monitoring. SFDA-aware validation and integration with HL7/FHIR systems.
Footfall, dwell time, queue management, and loss prevention. Integrates with POS for shrinkage analytics and conversion optimization.
Perimeter intrusion, vehicle classification, and target tracking — air-gapped, classified-grade. GAMI-aligned localization and integration.
Crop health, livestock monitoring, and yield estimation from drone and ground cameras. Designed for Saudi conditions: heat, dust, and scale.
Edge inference on Jetson, server inference on DGX, and a managed observability layer across both.
Walk the site. Pick three camera angles. Run a 2-week PoC against your real data and measure precision/recall before committing to anything bigger.
5–10 cameras, end-to-end. Edge devices commissioned, dashboards wired, alerting tuned to your operations team's tolerance for false positives.
Hardening, fleet management via NVIDIA Fleet Command, on-site commissioning, and integration with your incident management system.
Model retraining as conditions change (lighting, seasonality, new SKUs). 24×7 ops desk for incident triage. SLA-backed.
Faces, plates, and identifiable signals are processed and discarded at the edge. Only events leave the device.
Crowd-density estimation in real time. Incident prediction across the holy sites. Anonymized at source.
Radiology triage, ICU patient observation, and OR workflow analytics under SFDA validation.
Branch flow analytics, queue management, and physical security with SAMA-aligned governance.
Edge by default, on Jetson devices commissioned at each camera or each rack. Server-side inference on DGX is reserved for batch retraining and deep analytics. Live video never leaves the edge unless a workflow explicitly demands it.
Faces, plates, and other identifiers are blurred or hashed at the edge before any frame is stored or transmitted. Audit logs are PDPL-compliant. We've built clinical and public-space deployments under formal Privacy Impact Assessments.
Vertical-dependent. Person detection: 95–98% mAP on indoor scenes. PPE: 92–96%. Defect detection: highly task-specific — we measure on your data during PoC and contract to that number.
Continuous: edge devices receive signed model updates via NVIDIA Fleet Command. Updates are gated by canary deployment and automatic rollback on accuracy regression.
Yes — we calibrate against Saudi conditions specifically. PoC includes worst-case lighting and weather windows, and models are augmented for glare, haze, and night IR.
No. CCTV records and reviews. Computer vision detects, decides, and triggers — in real time, without a human in the loop for low-stakes events. The economic model is fundamentally different.
Thirty-minute working session with our Computer Vision lead. Bring three camera angles; we'll show you what they could be telling you.