Landslide
Slope failure onto the road or track.
Geohazard Early Warning
AI Cameras That Spot Landslides and Rockfall
An AI computer vision camera that recognises slope, road and bridge hazards on the device, day and night, in under a second.
What Does the Camera Detect?
Landslides, rockfall, debris flow and ground subsidence, plus structural failure such as bridge collapse.
Rockfall
Individual rocks found and outlined.
Debris Flow
Moving mud and debris, including early signs.
Bridge Collapse
Structural failure and abnormal movement.
Why Does Slope Monitoring Fall Short?
Patrols
Mountain roads and bridges are costly to walk and hard to cover often.
> 30%
Missed events can exceed 30% with single sensors, which also react to wind, light and vehicles.
> 5 min
Cloud analysis over weak mountain networks can delay a warning by more than five minutes.
How Does a Warning Reach the Road?
The AI runs on the camera, so a warning does not depend on the network.
- SeeThe camera watches the slope, bridge or portal, up to 400 m away.
- DecideThe edge AI chip recognises the hazard on the device, with recognition in under a second.
- Warn LocallyA wired roadside warning sign alerts drivers with no network needed.
- Inform the TeamAlerts, photos and live video reach the cloud platform and app over 4G.
What Is Inside the Camera?
F1.0 Super Aperture
Takes in about four times more light.
1/1.79 in Back-Lit CMOS
Nearly 50% more light than a standard sensor.
Uncooled Thermal Detector
Vanadium oxide focal plane with fine detail.
Edge AI Chip
High compute at low power on the camera.
Multimodal AI Model
Trained on images and text together.
Image Enhancement
Night boost, stabilisation and defog.
Moving Target Recognition
Filters static false alarms.
Multi-Band 4G
Stronger signal in remote places.
Cloud Platform and Mobile App
100,000+
Concurrent connections on a single server.
300 ms
Video latency, as low as. Manage cameras and alerts, and see alert photos and live video from anywhere.
Clear Pictures in the Worst Conditions
Drag each picture to compare the raw frame with the AI enhanced frame.
Night Enhancement
Defog
Stabilisation
Rain Removal
Why a Multimodal Transformer Model?
Hazard footage is rare, so the model learns from far less data and tunes in one step on a small set of site samples. Figures are the manufacturer's stated comparison with a CNN approach, with under 1% accuracy loss after quantisation.
-90%
Training Data Needed
-80%
Development Time
+50%
Accuracy in New Scenes
-50%
Compute Cost
| Compare | Multimodal AI Camera | Single-Mode AI Camera |
|---|---|---|
| Data | Image, infrared, audio and mmWave fused, such as visible light with thermal | Visible light only |
| Conditions | Handles low light and occlusion | Noise at night, overexposure |
| Accuracy | Features cross-check each other | Fooled by similar objects and angles |
| Decisions | Evidence from several sources | Vision only, no context |
Where Does It Protect People?
Mountain Roads
Watches for debris flow and rockfall beside the carriageway.
Railways
Early warning reduces injuries and property loss on the line.
Tunnel Portals
Finds slope risk above portals before it reaches traffic.
Bridges
Spots abnormal movement from load, weather or ageing.
Field Examples
Example alerts from the camera manufacturer's field use, not VYROX projects. Site details are withheld, and times run from the first alert.
Slope Failure Caught Live
The camera flagged the failed hillside, logged the event as debris flow, and held the alert open until an operator cleared it.
Bridge Collapse: Four Alerts in Two Minutes
First AlertRoad Collapse
+10 SecondsRoad Collapse
+45 SecondsAbnormal
+1 Min 47 SecAbnormalAlert after alert kept the failed span flagged, each with a photo.
Rockfall Across a Mountain Road
Four steady alerts in 40 seconds, each with a photo, confirmed that debris and rock were blocking the road.
Specifications
From the manufacturer's product data. The standard algorithms cover landslide, rockfall and debris flow; bridge collapse appears in field examples. This is a safety aid, not a guarantee: misses and false alarms can happen.
< 1 s
Recognition Response
400 m
Maximum Monitoring Range
> 90%
Recognition Accuracy
0.0001 lux
Full-Colour Night Vision
Imaging
- Resolution
- 4 MP, 30 FPS
- Sensor
- 1/1.79 in CMOS
- Aperture
- F1.0
- Thermal
- Uncooled detector
- Lens
- Zoom, sized to site
- Pan and Tilt
- Optional
- Dynamic Range
- Digital WDR
Connectivity
- Network
- Multi-band 4G and wired
- Protocols
- HTTP, RTSP, HLS
- Storage
- Up to 256 GB card
- Remote
- Live view, download, OTA, maintenance
Rugged Build
- Housing
- Aluminium alloy
- Protection
- IP67
- Temperature
- -25 to +60 °C
- Power
- 680 mA at DC 12 V
Questions
What is an AI geohazard camera?
It is an outdoor camera with an AI chip inside that recognises landslides, rockfall, debris flow and bridge collapse and raises an alert without sending video to a server first.
Does it work without internet?
Yes. Recognition runs on the camera and a wired local warning sign works with no network. 4G is used to send alerts and video to the cloud platform and app.
Does it work at night and in fog?
Yes. Night enhancement, defog, rain removal and stabilisation clean the picture, and it sees in full colour down to 0.0001 lux.
How fast is recognition?
Recognition response is under one second, compared with delays of minutes when analysis depends on the cloud.
Can it replace engineers or road closures?
No. It is a safety aid, not a guarantee. Misses and false alarms can happen, so it supports inspection and safety procedures and never replaces them.
Show Us One Camera. We Will Show You What AI Can See.
AI Vision Bots is available at selected sites. Send a short clip from your site and one sentence on the problem, and the VYROX AI team will tell you what a Detection could do there.