Skip to main content

The algorithm that detects crowds

Our new video analytics solution, Distributed Intelligent Video Analytics, helps passengers to avoid crowds.
It also provides transport operators with unique insights, alerts and predictions.


After a hard day at work, the last thing anyone wants is to squeeze into a packed metro carriage1 . Yet millions of us do it every day – or at least we did until

The pandemic has brought concerns about crowding on transport networks into sharp focus. The immediate need is to enable social distancing and to do everything possible to prevent the spread of coronavirus. 

Post pandemic, the ability to anticipate and manage crowding could be a differentiator. As well as easing passenger concerns, better crowd management has major operational benefits including reduced dwell times and improved service reliability.

So how can all of this be achieved? Easing the crowds with Distributed Intelligent Video Analytics

Better management of crowds starts with better data. And that’s exactly what Thales Distributed Intelligent Video Analytics provides. 

Distributed Intelligent Video Analytics works by extracting people counting from CCTV cameras using computer vision algorithms. This data is used to generate real-time heat maps and KPIs to provide insights for different users. 

The beauty of Distributed Intelligent Video Analytics is that it uses existing CCTV cameras to gather data. No calibration of cameras, nor new sensors are required. Cameras can be anywhere – on stations, on platforms and on moving trains. 

The software behind Distributed Intelligent Video Analytics can be delivered by Thales as a digital service through a distributed architecture; the algorithm may be hosted on stations, or deployed on trains. The solution is futureproof, offers maximum flexibility and can be integrated with third-party applications.

Here are some of the ways that density data can help passengers and operators:

Passenger guidance on platforms
This is a powerful crowd-avoidance tool that shows passengers waiting on platforms which carriages of approaching trains are crowded and which are not. This information is shown on a platform display with carriages colour-coded according to density: green for low, yellow for medium, red for high. The result is faster embarkation and disembarkation, reduced dwell times, greater reliability and happier passengers. 

Live network density maps 
Staff in the Operation Control Centre (OCC) need a quick and easy way to see where there’s a crowding problem – without having to call up dozens of different cameras. Live network density maps provide instant insights: the passenger density level of each station is displayed, with a colour-coded indication for each station (green, yellow or red). Platforms and stations are permanently monitored and density thresholds can be pre-configured.

Live station density maps 
These provide OCC operators with a clear real-time indication of crowding levels on stations. Information is shown on a heat map, so you can instantly see which parts of the station require attention. Every corner of the station can be monitored. 

Intelligent crowd monitoring alerts
Our intelligent alerts provide an early warning about abnormal passenger density levels within a given area. The reference density level can be automatically correlated with historical data, based on location, date and time of day. This reduces nuisance alarms and makes it easy to detect abnormal events. 

Occupancy KPIs and trends
What percentage of my network is occupied by passengers? Are there more passengers than usual? Or fewer? Is passenger density rising? Or falling? Our passenger density dashboard provides operators with a global overview of their networks in real time.

Historical data
Traffic managers can easily compare historical data and network occupancy to identify emerging trends, understand passenger ridership and answer complex operational questions with confidence, so timetables can be optimised. 

Density predictions
Based on both current occupancy and historical data, these help operators to adapt train services to forecast demand. In the case of unattended CBTC metros, extra trains can be summoned from distant stabling points and launched into service at exactly the right moment so they reach the congested part of the network just as the crowds are starting to grow.

What’s next?

The technology behind Distributed Intelligent Video Analytics opens up a number of exciting new use cases. These build on the latest artificial intelligence (AI) techniques and include trespassing detection on platforms, unattended luggage detection and remaining passenger detection at the end of the line.


 1 - A study demonstrates that live information showing train occupancy levels is in the top 4 of “what passengers really want”: What passengers really want: Assessing the value of rail innovation to improve experiences, Luis Oliveira, Claudia Bruen, Stewart Birrell, Rebecca Cain - Transportation Research Interdisciplinary Perspectives, Volume 1, June 2019, 100014,