Challenge

Ensuring efficient and punctual train operations is a core priority for Infrabel, the organisation responsible for managing Belgium’s railway infrastructure. The rail network faces frequent operational challenges, including:

  • Unplanned delays caused by obstacles on tracks such as construction material, unauthorised persons walking on the rails, accidents at crossings, cable theft, etc.
  • Manual incident detection, which relies on human observation and is prone to delays and errors
  • Limited visibility on train locations, making it difficult to accurately track train movements across the network
  • Slow response times when disruptions occur, affecting overall punctuality and passenger satisfaction.

Without an automated system, Infrabel’s dispatchers and operators rely on camera feeds and manual reporting to assess train locations and potential disruptions. However, this approach lacks real-time situational awareness and requires constant human oversight, leading to inefficiencies in responsding to track obstructions, stalled trains, or operational issues?

Solution

Together with Infrabel, we developed an AI-powered tracking system that utilises computer vision and object recognition to monitor objects across Infrabel’s railway network. The scope of the Proof-of-Concept entailed detection trains and train numbers. By processing video footage from existing surveillance cameras, the system can:

  • Detect stationary and moving trains in real-time
  • Recognise train identification numbers using Optical Character Recognition (OCR)
  • Analyse traffic conditions and identify bottlenecks and disruptions
  • Provide insights to railway operators for improved decision-making

This automated detection and monitoring system reduces the need for manual supervision, ensuring faster response times and better train scheduling.

Approach

We executed the project in two phases, with phase 1 already developed and phase 2 set as the next step towards full deployment.

Phase 1: Proof-of-Concept

  1. Data collection and model training: video footag from Infrabel’s surveillance network was used to train a deep learning model for train detection
  2. Object recognition & OCR integration: the AI system was designed to detect trains and extract identification numbers from stationary and moving vehicles
  3. Prototype development: a basic application was developed to allow Infrabel to upload videos, analyse footage, and visualise detected train movements
  4. Accuracy testing & refinement: the model was tested across various lighting and weather conditions, ensuring robust performance in real-world scenarios

Phase 2: Expanding capabilities & deployment

  1. Enhancing video processing efficiency: upgrading the AI pipeline to handle larger video files more efficiently
  2. Integration of text detection & OCR as microservices: adding a dedicated module for recognising and extracting train identification numbers
  3. Scaling to real-time video streams: expanding from offline video analysis to live stream processing, enabling real-time monitoring.
  4. Performance benchmarking & validation: conducting a full-year analysis of train detection performance across different environmental conditions
  5. Preparing for full deployment: finalising system improvements before integrating with Infrabel’s existing railway traffic management infrastructure

By structuring the project into phases, Faktion and Infrabel ensure a scalable and reliable AI deployment that aligns with operational needs.

Outcome

The AI-powered train tracking system has demonstrated significant potential to improve railway operations:

  • Automated train monitoring: reducing reliance on manual monitoring, freeing up resources for higher-priority tasks
  • More accurate traffic insights: providing operators with real-time visibility into train movements
  • Improved incident response: faster detection of stalled or obstructed trains, allowing for quicker interventions
  • Scalability for future expansion: laying the groundwork for network-wide deployment in future phases.

While phase 1 validated the feasibility of AI-driven train tracking, phase 2 will focus on expanding and refining the system, integrating real-time capabilities, and preparing for full-scale deployment.

By applying computer vision for object recognition and tracking, Infrabel is taking a proactive, data-driven approach to managing railway traffic, improving punctuality, efficiency, and safety across Belgium’s rail network.