The video-AI spine
across the platform.
AeroVision is the video-AI capability NorthSky uses across every camera in the airport. It's the model family behind AeroGround's turnaround detection, behind AeroFOD's 1–2 cm debris detection, behind AeroRunway's touchdown and rollout capture — and behind the terminal-side passenger-flow, queue-heatmap, and incident-search work in active design-partner development. You don't buy AeroVision separately — you buy the product that uses it. This page tells you how the spine works.
Where AeroVision powers
the platform.
The same video-AI capability runs at the stand, at the runway edge, on the mobile FOD vehicle, and (in development) in the terminal. Each product uses a tuned version of the model for its operating environment, but they share the same training pipeline, the same edge runtime, and the same governance. When a model improves for one product, the others gain.
AeroGround.
Detects every turnaround event from the apron camera frame — chocks, aerobridge connect/disconnect, fuel-bowser arrival/depart, catering, baggage loading/unloading, cleaning, pushback. Plus apron FOD, GSE positioning, and ramp-safety violations. Fused with stand sensors; structured events onto the ATOMS bus.
AeroFOD.
Detects 1–2 cm foreign objects across every paved surface of the airfield — mobile vehicle, fixed pole-mounted units, and stand-mounted units. Classifies object type (metal, rubber, fluid, biological); estimates size; emits coordinates to the safety dashboard.
AeroRunway.
Captures touchdown, rollout, RET vacate, line-up, take-off roll from the runway-end masts. Confirms ROT per movement; flags runway incursion risk; produces the visual evidence behind the configuration and sequencing advisories. Pairs with surface-radar and MLAT.
Terminal intelligence.
Passenger flow, queue heatmaps, gate-area density, signage and PA coverage, and a searchable incident-video record across every camera in the terminal. In active development with selected design partners on terminal-heavy airports. Privacy-by-design — counts and densities, no biometric identification.
What the spine does.
AeroVision is a stack of model families tuned for the airport operating environment — apron lighting, monsoon visibility, runway distance, terminal crowd density. Each family is independently trained, independently versioned, and independently governed. Together they cover what an airport needs cameras to do.
What's in the frame.
Detects and classifies every object in the camera frame — aircraft, GSE, vehicles, crew, baggage carts, fuel bowsers, FOD. The foundation model behind AeroGround's event capture and AeroFOD's debris detection. Tuned per camera angle, per stand type, per surface.
- Vehicle and aircraft classification across narrow-body, widebody, freighter.
- Crew and PPE detection for ramp-safety compliance.
- FOD classification across 14 classes — metal fastener, rubber, plastic, fluid, biological, ice/snow.
What's happening.
Recognises actions and state transitions — chock placement, aerobridge connection, door open/close, fuel hose connect, pushback commence. The temporal model behind the turnaround event catalog. Trained on the operational procedure book; configurable per airline.
- 15+ turnaround events per stand, per turn.
- Time-stamped at the edge to the platform clock.
- Confidence band per event — low-confidence routes to supervisor review.
When the camera isn't sure.
Fuses vision with non-camera signals — weight-on-wheels, aerobridge state, fuel-flow signal, mm-wave radar for FOD, multilateration for tail attribution. The confidence layer that lets us claim the operating record reconciles by construction.
- WoW + aerobridge sensor confirms arrival anchoring.
- 77 GHz mm-wave radar fusion for all-weather FOD detection.
- MLAT + ADS-B attribution per detection to a specific tail.
Every camera, searchable.
Cross-camera, time-indexed video record across the whole airport — terminal, apron, taxiway, runway. Search by time, location, flight, or event type. Evidence chain-of-custody preserved for safety review, insurer claims, grievance redressal, and regulator inquiries.
- Search by stand, time, flight, event type, or natural-language query.
- Configurable retention per camera per zone.
- AAIB evidence pack auto-assembled on a single event ID.
Where the queues form.
Continuous passenger-flow analytics from terminal cameras — security lanes, immigration, boarding gates, baggage reclaim. Surfaces bottlenecks in real time and after the fact. Privacy-by-design: counts and densities, no biometric identification.
- Queue length and wait-time estimates by location.
- Per-hour, per-terminal heatmaps for operations review.
- Correlation with AeroFeedback — sentiment by queue point.
Boarding readiness, visible.
Detects crowding around gates that signals a boarding call is imminent, or a slip ahead. Helps ops staff anticipate which gates need PA attention, which need extra staff, which are likely to miss schedule. Pairs with the ATOMS turnaround stream.
- Per-gate occupancy trend over the boarding window.
- Operations record alongside flight events.
- Triggers contextual surveys when integrated with AeroFeedback Operational Triggers.
Terminal-side AeroVision is in design-partner phase.
Passenger flow, queue heatmaps, gate-area density, and the cross-camera incident-search experience are in active development with selected design partners on terminal-heavy airports. Four weeks of CCTV access, two hours per week from a named operations lead, joint case study published only with your approval, pricing reflecting partnership status through the roadmap window.
Buy the product.
Get the spine.
AeroVision isn't a SKU — it's the video-AI spine that comes with AeroGround, AeroFOD, AeroRunway, and the terminal-intelligence work in design-partner phase. Start with the product that fits your operating priority; the spine comes with it.