Explore ground truth for rail

We map the future of rail transformation. Our ground truth annotation platform is perfectly equipped to support a wide range of rail classes and attributes.

We lay the tracks for rail innovation

Our ground truth annotation platform combines cutting-edge machine vision and automatic perception technology to develop reliable autonomous rail applications. With advanced labeling technology, we ensure high-quality, scalable ground truth rail annotations. Discover the infinite possibilities of innovation in the rail industry.

Ground Truth use cases within rail

From Automatic Train Operations (ATO) to safety monitoring within railway stations, understand.ai effectively addresses all relevant rail use cases.

2D Driver Monitoring System

Automatic Train Operations (ATO)

Ground truth data facilitates precise monitoring and control of train movements, ensuring safe and efficient operations without direct human intervention, enhancing rail network capacity and reliability.

2D Driver Monitoring System

Monitoring of rail infrastructure

Leveraging ground truth insights, rail infrastructure such as tracks, signals, and bridges is continuously monitored for defects or anomalies, enabling timely maintenance and minimizing disruptions to train services.

2D Driver Monitoring System

Safety in railway stations

Ground truth information supports safety monitoring and on-offboarding processes at railway stations, enhancing customer experience and ensuring passenger safety during arrivals and departures.

2D Driver Monitoring System

Rail warehouse maintenance

Improving inventory management efficiency and reducing risks, ground truth data supports in the upkeep of railway warehouses, ensuring optimal conditions for storing and handling goods.

    Reference Case: European Rail Operator

    Explore this following reference case to gain deeper insight of what a successful project with us could look like.

    In Germany, Deutsche Bahn's Rail Infrastructure Manager (DB InfraGO) has set-up the Digitale Schiene Deutschland sector initiative (DSD) to pursue a high degree of automation and digitization of railroad operations. In the context of fully train automated train driving (so-called grade of automation level 4, GoA4), sensors and cameras are needed to be able to react automatically to risks and dangerous events in the rail environment by means of artificial intelligence (AI); to this purpose lidar and radar sensors as well as cameras are used to automatically detect and respond to hazards, such as objects on the track or vegetation or irregular situations nearby, or passengers in stations in dangerous proximity of the track. Another important use case is high-precision train localization by detecting static infrastructure elements and landmarks and locating them on a digital map.
    To develop such a AI software functions for environment perception, very large amounts of data, recorded with very high quality under different conditions and scenarios, are required. In this way, tests and simulations can be carried out and AI models can be validated, certified and used in automated railroad operations.

    The Challenge

    • Develop and engineer a high throughput/high quality data labeling service able to deliver consistent and predictable quality and the flexibility to adapt to changes of Data formats, sensors and labeling specifications, in order to support Rail Automation.

    Key Facts

    • Object catalogue and labeling specs include 50 rail objects’ classes and each object has 6 attributes on average;
    • Sensor Fusion includes multiple sensor data formats and datasets : Infrared , long range and mid-range cameras, lidar, radar
    • Flexibility of data input: Implementation of new or evolved sensor set-up every 5-6 months and adaptation of Labeling Specifications

    Key Achievements

    • Upfront quality control over raw data were able to identify sensor calibration and frame synchronization issues, before data labeling to avoid rework and delays.
    • Delivery throughput = average throughput 110k annotations/week (corresponding to 10 clips/week), with peak throughput of 140k annotations/week.
    • Delivery quality/ data labeling accuracy: +99% on average during the period
    • Final quality rate: for the delivery of 194 clips and +2M annotations, we achieved 1st time customer acceptance rate of 100%