THE CONTEXT
During wildfire season, strategic airstrips located deep in the forests are critical infrastructure: water-bomber aircraft operate from them, loading water to fight fire outbreaks. Every minute counts and every operation must be recorded. Before this project, control relied on manual logs: which aircraft arrived, at what time, how long it stayed on the strip, whether it actually loaded water, and whether ground personnel complied with the required personal protective equipment. There was no objective traceability and no real-time information to coordinate operations or audit air operators.
THE DEPLOYED SOLUTION
We designed, installed and operate a computer vision platform that turns each airstrip's cameras into business sensors, operating autonomously and continuously, with on-premise processing:
- Automatic logging of arrivals and departures: the platform detects each aircraft on live video, tracks it frame by frame throughout the operation and records arrival time, departure time and exact time on the strip, with no human intervention.
- Automatic aircraft identification through industrial OCR of the registration painted on the fuselage and wings, with a rescue algorithm that corrects typical reading confusions and evidence photos for human verification.
- Automatic detection of the water loading process: the system recognizes when the aircraft connects the refill hose and measures the exact duration of the process, distinguishing an operational landing from a simple stopover.
- Complete audiovisual evidence for every event: up to twelve photographs per operation (arrival, departure, registration, water and fuel loading) plus event video, available on the web for review and auditing.
FUEL LOADING MONITORING
The platform also supervises the aircraft refueling process on the strip, a critical point of the operation both for safety and for service traceability:
- Automatic detection of fuel hose connection and disconnection, recording start time, end time and exact duration of each refueling.
- Photographic evidence of the process linked to the identified aircraft, allowing each refueling to be audited against the fuel supplier's records.
- Consolidated log per aircraft and per period: number of refuelings, average loading time and automatic cross-referencing with water-loading operations, to reconstruct each aircraft's full cycle on the strip (arrival, fuel, water, departure).
SAFETY COMPLIANCE MONITORING (PPE)
A second AI model verifies in real time that personnel in the loading area wear a safety helmet and a reflective vest, generating evidence for every non-compliance and turning safety compliance into a measurable, auditable indicator:
- Detection of safety helmets and reflective vests on personnel in the operation area.
- Compliance indicator panel: detections analyzed, non-compliance rate, missing helmet and missing vest per period.
- Downloadable PDF compliance report by date range.
THE PLATFORM
- Real-time operational view with aircraft currently on the strip, daily indicators and the latest detections with their evidence.
- Cameras grouped by airstrip with online/offline status and live visualization of AI processing on each camera.
- Interactive dashboards: operations per period, registration reading success rate, average time on the strip, aircraft ranking, hourly distribution and activity heatmaps by day and hour.
- Executive PDF reporting generated directly from the platform: operations report per aerodrome, report broken down by aircraft and PPE compliance report, plus data export for external analysis.
- Full administration panel to review, correct and audit every record, and an integration API for external systems.
CONTINUOUS MODEL IMPROVEMENT
The models were trained on thousands of real operation images, curated and labeled through our own data pipeline: the aircraft and people detector reaches 98.7% precision (mAP50) and 99.0% recall with 4,351 training images —including daytime, night-time and low-visibility scenes— and the PPE compliance model reaches 95.7% precision with 10,756 real images of loading areas.
RESULTS
- Autonomous, continuous monitoring of multiple forestry airstrips from a single centralized platform, without permanent human surveillance.
- Complete traceability of every air operation: identified aircraft, arrival and departure times, time on the strip, and photographic and video evidence.
- Objective control of the service: automatic water and fuel loading detection verifies that every operation fulfilled its purpose.
- Improved safety standard: automatic PPE monitoring that is measurable and auditable.
- Executive reporting ready for auditing, billing and decision-making during wildfire season.
- Scalable platform: adding a new airstrip is an installation and configuration process, not a new development.
Real data, real results