An early-warning system for birds and drones near runways
SkySentry turns an ordinary phone into a lookout that spots birds and drones near a runway and alerts the control tower, out loud and on screen, before a plane takes off into danger.
- Detects
- Birds with the camera, drones with the microphone
- Alerts
- Spoken and on-screen alerts in the tower within seconds
- Built for
- Small airports that can’t afford bird radar
- Costs
- A phone per runway plus free-tier cloud tools
SkySentry
hackPHS 2026 · Into the skiesSkySentry
An early-warning system for birds and drones near runways
Geese knocked out both engines of US Airways Flight 1549 minutes after takeoff.
Drone sightings shut down Gatwick Airport over three days.
Radar is for big airports
- 1Detect and send
- 2Save the stream
- 3Verify with AI
- 4Store the report
- 5Stamp drone reports
- 6Give it a voice
- 7Tell the tower
SkySentry
An extra set of eyes for the airports that can’t afford a million-dollar system.
BUILT IN 24 HOURS · HACKPHS 2026 ElevenLabs · Gemini · Tiger Data · DigitalOcean · MongoDB Atlas · SolanaSpace to play or pause · arrow keys to change scene
Scene details & asset credits
The original eight-scene explainer, enhanced with a detailed airliner, PBR ground materials, natural environment lighting and a more fully dressed airfield. This is an illustrated prototype demonstration; detection readings, risk events, reports and charts are simulated.
- Narrow-body airliner — Innerscene, CC0. Scaled to the original animation; animated fans and retracting landing gear.
- Clean asphalt — Dimitrios Savva / Poly Haven, CC0. Color, normal and roughness maps.
- Grass ground — Charlotte Baglioni / Poly Haven, CC0.
- Derelict Airfield 01 — Alexander Scholten / Poly Haven, CC0. Used for reflected light.
- Three.js — MIT. Cinematic pass (physical sky, ambient occlusion, bloom, depth of field, film grade, lens flare) uses Three.js locally bundled add-ons; clouds, grass, farmland, village, navigation aids and particles are generated in code.
- Scanned props from Poly Haven, CC0: shrubs, boulder, street lamp, security camera, utility box, air-conditioning unit, concrete road barrier, portable generator, LPG tank and fire hydrant. Authors and checksums are in the asset manifest. Full asset manifest.
Airfield layout, captions, detection sequence, risk rules, architecture and 100-second timeline retained from your supplied HTML. The scenery is illustrative. Historical context: NTSB / Flight 1549.
Illustrated prototype. Detection readings, alerts, reports and charts in the film are simulated.
The sky over a runway is crowded
Birds and drones share the same low airspace that every departing aircraft climbs through. Big airports pay for radar to watch it. Most small airports watch it by eye.
engines lost when geese struck US Airways Flight 1549 minutes after takeoff, forcing the “Miracle on the Hudson” landing.
flights disrupted when repeated drone sightings shut down London’s Gatwick Airport over three days.
at small airports like Princeton and Trenton-Mercer, which usually can’t buy bird radar or counter-drone systems.
It sees, it hears, it speaks up
One phone on a tripod at the runway edge does the watching. A second laptop in the tower shows and speaks the alerts.
Sees birds
A ready-made AI model runs in the phone’s browser and boxes every bird. Tracking counts them and shows which way each one is heading.
TensorFlow.js · COCO-SSDHears drones
Drones are hard to see but easy to hear. A sound model trained on background noise, drones and birds listens for the buzz.
Teachable Machine audioTells the tower
Risky events reach the tower screen in seconds with a spoken alert, an AI-written incident report and a pin on the airport map.
Socket.IO · ElevenLabs · GeminiFrom the runway edge to the tower
A phone mounted at the runway watches the sky, and the tower hears about it in under a second.
Three levels, one clear rule each
The phone scores every moment as green, yellow or red, so the tower only hears about what matters.
No birds, no drone sound
Quiet status update every few seconds
1 to 4 birds, not heading toward the runway
On-screen caution, snapshot checked by Gemini, logged
5 or more birds, any bird heading toward the runway, or any drone sound
Alarm, spoken alert, Gemini report and map pin; drone reports also get a Solana stamp
How one alert travels
Every alert follows the same seven steps from the runway to the tower. Each service does one job.
What it can’t do yet
- It works in daylight only. Night would need a thermal camera.
- The off-the-shelf model works best when birds are reasonably large in the frame.
- It is an extra set of eyes for the tower, not a replacement for radar.
- Camera stationDetect and sendThe phone spots birds or a drone, scores the risk and sends an event with type, count, direction and confidence.
- Tiger DataSave the streamEvery event is stored with a timestamp, and hourly summaries feed the tower’s charts.
- GeminiVerify with AIFor yellow and red events, Gemini checks the snapshot and writes a short incident report.
- MongoDB AtlasStore the reportThe report, snapshot and Gemini’s label are saved together.
- Solana devnetStamp drone reportsA hash of each drone report goes on chain, so nobody can quietly edit it later.
- ElevenLabsGive it a voiceThe alert sentence becomes a spoken tower announcement.
- Socket.IOTell the towerThe alert, report and audio arrive on the tower dashboard instantly.
Run SkySentry right here
Open the tower on one screen and a camera on another. Detections travel through this server to every tower at the same airport in under a second.
Camera station
Point a phone or laptop at the sky. Vivan’s engine finds birds by motion, counts wingbeats, groups flocks and works out how many seconds until they reach the runway.
Webcam · video clip · synthetic sky
Tower dashboard
Live risk, spoken alerts, an airport map with each threat’s bearing, hourly charts, AI incident reports and a full-screen software beacon. No extra hardware.
Socket.IO · Tiger Data · MongoDB · Gemini
Engine tests
Run the detection engine’s test suite and scenario accuracy table in your browser: flocks, lone birds, drones, plastic bags and planes.
Synthetic scenarios · accuracy table
Free tools, each doing a real job
What people ask
How accurate is it?
The team measures it on a set of test clips before judging and reports the result, for example how many of 20 clips it caught. Treat it as an extra set of eyes, not a replacement for radar.
Can it see small, far-away birds?
The free model works best when birds are reasonably large in the frame. A zoom lens or a model trained on sky footage would improve that.
Does it work at night?
Not yet. Night detection would need a thermal camera, which is a natural next step.
What about privacy?
The camera points at the sky, and snapshots are saved only when there’s an alert.
Why use a blockchain?
So drone incident records can’t be quietly changed later, which matters if they are used as evidence for police or the FAA.
Isn’t this already done?
Big airports use radar and commercial counter-drone systems. SkySentry is a low-cost version for the small airports that can’t afford those.
What would it cost a real airport?
A phone or cheap camera per runway, plus small cloud bills.