Zircon Presents: PTI Cam Software Solutions

Over the course of the last few years, the rail industry has seen a renewed focus on the safety of passengers at the Platform Train Interface (PTI). It may sound surprising but incidents at the PTI are one of the biggest contributors to passenger injury and fatality rates. 2015/16 saw five passenger fatalities in the UK alone.

It seems that, despite ongoing attempts to raise public awareness, regular platform announcement and the yellow line, there is still a degree of naivety that by getting too close to the edge of a platform or venturing into the track well, passengers open themselves up to a number of dangers.

In the hopes of negating the frequency of incidents, Zircon have conducted some early concept work into software solutions utilising video analysis techniques aiming to improve the level of safety at the PTI through the real-time analysis of CCTV footage.

As things stand, we have been able to investigate the feasibility of using software to detect just two of the potential incidents that are frequently witnessed at the PTI. These are Drag Detection and Track Incursion.

Drag Detection

When a passenger become entrapped in the doors of a train, either by their clothes or by belongings such as bags and suitcases, or between the train and the platform there is a high likelihood that they will find themselves being dragged along as the train pulls away. Should the incident go unnoticed, the individual being dragged faces a very real risk of serious injury and in certain instances death.

If you add the fact that the number of passenger journeys being made is continuing to increase each year, and with it the frequency of overcrowding at platforms (especially around peak hours and larger cities), we can expect to see a concerning change in the number of passenger drag cases.

With this in mind, we decided to see if it would be feasible to design a system capable of automatically identifying a situation where a person or object is being dragged along a platform by a departing train.

It was our aim to design a system that would be able to react instantaneously and raise an alert to either the train driver or platform staff, so as to prompt a more rapid reaction and reduce the chance for injury.

“Customers are reminded to mind the gap between the train and the platform edge.”

Demonstration of Drag Detection in Action During an Incident in Rome

Unfortunately, due to the nature of the technology, it is hard to conduct real-world, full-scale tests, and the likelihood of witnessing such an event during a specific testing period is very low. In order to work around these difficulties, we have had to manage by running a number of recordings from various stations and sites, where these incidents have occurred. One such incident captured recently, by CCTV cameras at a metro station in Rome, caught our eye after it was reported by the BBC.

With the incident being a realistic real-life event we were curious to see what would happen if we were to run it through our Drag Detection software, and as you can see from the video above, the outcomes from the test were very positive.

Of course, no amount of testing in an artificial environment that does not take into account all of the potential conditions, flaws or situations, will provide definitive proof that our solution is efficient and effective.

Track Incursion

Hearing the term Track Incursion is bound to conjure images of obvious dangers such as passengers climbing or jumping down onto the track, however even something as simple as leaning over the platform edge can put passengers at serious risk. With the rapid increase in mobile devices, there has been a worrying increase in the number of incidents where passengers are encroaching into the operational track space whilst attempting to retrieve dropped devices.

After plenty of internal research, we had identified a potential solution that could identify track incursion events from CCTV footage using computer vision techniques. Curious to find out the degree of successful detection these techniques could achieve, we performed a number of trials here at our office in Trowbridge and were pleased to see successful results.

, Of course, there will be a difference between detecting an incursion from the pre-recorded test footage we used for our internal trials and achieving the same results in real-time in the environment of a functioning railway platform.

Having been fortunate enough to receive funding from the Department for Transport’s (DfT) Transport Technology Research Innovation Grant (T-TRIG), with assistance from London Underground we were able to collect footage from cameras installed at a platform on the Victoria Line over a 12 week period.

Example test of Zircon’s Track Incursion PTI Cam solution

So that we could prove the viability of our solution in a commercial environment we decided to analyse functionality against the following criteria:

  • System Costs: What are the commercial prices of installing and supporting the infrastructure that will allow the techniques to be used in a viable way for the railway?
  • Accuracy: How often does it produce false negatives/positive? How does this compare to true negatives/positives?
  • Duration: How long must an Incursion be in frame before the method detects it?
  • Distance: How far away from the camera, must an incursion before it is not detected?
  • Size: How small of an incursion (i.e. whole body, half body, arm only, hand only) must be made into the incursion area before it is not detected?
  • Sustainable: How well does our hardware function in a real-world environment, and what is required to support the infrastructure that has been installed?

Sadly during our trial period in the Underground, there was an incident where a passenger collided with a moving train. Whilst these kind of events aren’t pleasant to even consider let alone watch, we were provided with the opportunity to analyse the behaviour of this passenger with the hope that we can find a strategy that will highlight other passengers expressing similar behaviours and intervene before an incident is able to occur.

“The risk at the PTI accounts for 36% of the total fatality risk for passengers” – Department for Transport Rail Accidents and Safety Statistics

Unfortunately, whilst the technique being trialled was able to detect the presence of the passenger and the difference in the speed and stopping position of the train as the driver attempts to stop before the collision can occur, the incident simply happens to fast for it to be feasible to raise an alert. However, we did notice that the way the passenger behaves was markedly different from that of a typical passenger e.g. arriving two trains before the collision event, nervously pacing when the rest of the crowd were stationary and venturing dangerously close to the track to plan their jump.

Although we did conduct a basic investigation into the possibility of using Optical Flow methods to detect ‘high risk’ passengers we didn’t delve too deep, as although relevant to our investigation developing a new method fell outside of the scope of the project.

Whilst we are still looking into the results of the trial, early conclusions have found that our solution is able to function efficiently and consistently within the test conditions. As you can expect there are still some improvements to be made, however, we are now in a position that will allow us to move forwards with the addition of a new potential avenue of investigation.

Future Plans

As you can expect there are still some improvements to be made, and both of these solutions are still little more than a drop in the ocean. However, we are in a position that will allow us to continue to move forwards, especially having found a new potential avenue of investigation.

“In the interest of safety, please stand back from the platform edge.”

Like What You See and Looking For More?

If you feel that our experiences of utilising Video Analytic techniques to improve passenger safety at the PTI are of interest to you, or you would like to know more about the technologies under investigation, we would love to hear from you.