Smarter rail safety at the edge
_{Capgemini and Qualcomm are making railway crossings safer}

When a car breaks down on a railway crossing, every second counts. A fast-moving freight train might need over a mile to stop, and even a few seconds ’delay in alerting the driver could mean the difference between a safe rescue and a catastrophic collision.

To reduce that risk, Capgemini Engineering teamed up with Qualcomm Technologies, Inc. to explore how artificial intelligence (AI) can help. The result is a smarter way to monitor rail crossings – using powerful, low-power AI chips embedded at the edge of the railway network.

The rail crossing safety problem – AI to the rescue

Railway operators are under constant pressure to make crossings safer. In the United States alone, incidents at highway-rail grade crossings occur around 2,000 times a year. These incidents are not only dangerous – over 40% involve injuries or fatalities – but expensive, disruptive, and difficult to prevent.

Traditionally, detecting a vehicle stuck on the tracks has involved bulky, centralized systems that rely on cloud computing. They’re often slow to process alerts, rely on constant connectivity, and can be expensive to scale or update.

Working with an American Class-1 freight railroad client, Capgemini set out to change that. We developed and trained an AI-powered visual monitoring system that uses cameras and machine learning to spot potential dangers in real-time. But to make the system faster, more efficient, and widely deployable, we needed smart hardware.

Enter Qualcomm’s chips – turning AI models into physical, scalable products

That’s where the AI enabled Qualcomm Dragonwing™ QCS6490 processor comes in. Part of the broader Dragonwing^TM portfolio, it advances intelligence at the edge—delivering efficient, high-performance compute and on-device AI processing with advanced connectivity to transform industrial systems.

Capgemini integrated its AI software into the Inventec AIM-Edge QC01, a compact edge AI device powered by the Qualcomm processor.

This brought several improvements.

First, it dramatically reduced the system’s reliance on the cloud. Instead of constantly sending video footage to distant servers for analysis, the AI now runs directly on the device, right at the crossing. That means faster detection, quicker alerts, and fewer chances for network lag to interfere.

Second, the chip’s built-in AI processor – a neural processing unit, or NPU – makes the whole system more efficient. AI analysis that once taxed the device’s memory and slowed performance now runs smoothly, using 33% less memory and 5% less CPU power, all while making AI decisions in just 18 milliseconds per video frame.

Third, the solution can scale. Thanks to support for up to five simultaneous camera feeds, the same system can be adapted for different safety scenarios – not just crossings, but stations, tunnels, and even inside trains.

All of this required some customization of Capgemini’s original AI model to take full advantage of Qualcomm’s dedicated AI hardware. There was no need to retrain the model, but deep technical work was required to convert it into a format optimized for the Qualcomm NPU, and then to fine-tune it for the new setup.

Why edge AI matters for rail

For rail operators, this kind of edge AI is a practical solution to a longstanding problem that centralized IT systems never quite solved. It’s cheaper, because it reduces cloud usage. It’s faster, because it processes information locally. And it’s more versatile, with the ability to scale and evolve to different scenarios.

Capgemini estimates that performing the video analytics on the edge AI device reduces the total cost of the solution by 30% vs a cloud based alternative.

Perhaps most importantly, it opens the door to rapid innovation. Once we had integrated our initial rail crossing model, Capgemini was able to build and deploy new applications into the model – including the detection of weapons and violent behavior – in just a few days.

For industries like transportation, logistics, and infrastructure, this shift to the edge is transformational. It allows organizations to respond to real-time events, manage operations more efficiently, and improve safety without relying on massive data centers or always-on internet connections.

What’s next?

Capgemini is now preparing to roll out its Qualcomm-powered monitoring system in live rail environments.

The technology is expected to be deployed in crossings, stations, and other high-risk areas, creating a smarter, more responsive safety net across the rail network. And with scalable platforms like Qualcomm’s Dragonwing™, the journey from prototype to production is faster and more seamless than ever.

For more information

Contact Capgemini Engineering to learn more about our work in the rail sector or read our vision for the rail sector: Rethinking Rail – The Digital Transformation in Railways.

A detailed technical description of this project by experts at Capgemini and Qualcomm is available here: Capgemini leverages Qualcomm Dragonwing portfolio to enhance railway monitoring with Edge AI.