Mobile Monitoring

Train-borne vision systems
for ongoing monitoring

Our engineering, AI, and mathematics teams advance systems that visualise and contextualise large sets of geosensing, lidar, and camera data.

  • Automated Track Inspections
  • Automated Track Geometry
  • Rail Profile
  • Ride Quality
  • Positioning
  • 70,000 frames a second
    up to 200 km/h; 0.8mm resolution
  • 90%+ of data reduced
    by using edge computing in vehicle
Edge Processing

We’ve integrated edge and deep learning with our vision systems for near real-time anomaly detection.

We’ve converted vision systems to edge—where beneficial. Track data that’s too large to send wirelessly is shipped to us to identify potential defects.

We alert track managers to anomalies.

Case study:
Hitachi

Not only used for dedicated inspection trains, our systems now collect data from passenger trains

Track geometry systems on:

  • 16 IEP trains
  • 4 Abellio EMR trains
Case study:
Network Rail

With 15 years of stored data, we can determine change detection across longer timespans using AI

  • 25TB of images per day from each of the five inspection trains (10TB per 700km)
  • Each train scans 185,000km of track/year
  • 300 track inspectors spend less time in trackside positions of risk

GPUs
+ Data Centres

Our vision systems use edge GPUs to process data on the train in real-time. Or move it to our/your data centre—to post-process with AI.

Faster. Safer.

OmniView is our cloud platform and data historian, where clients can access all their automated inspections and alerts.

What's evolved in the last few years?

Edge hardware + 5G replaces Solid State Drive (SSD) shipments. Deep Learning replaces traditional machine vision.

Business Continuity + Safety

Designed to withstand security breaches or natural disasters, our data centres gain 25Tb+ of data everyday, continuously training our models.

We've started fusing data in a way not possible in 2020.

By combining multiple data sources over longer periods of time, we’ve been able to pre-empt anomalies hidden when observing just one data source.

Lianne Crooks
Head of Data Science & Maths, Omnicom Balfour Beatty