Human behaviour modelling in evacuation scenarios – Project Getaway
Published: 08 June, 2012
Did you know that only 38% of people register the presence of an emergency exit sign during an emergency evacuation? IFJ talks to Professor Ed Galea, evacuation modelling specialist at the University of Greenwich, about Project Getaway and the latest developments in evacuation.
Human behaviour in emergency evacuation scenarios is a profoundly complex area and one that is constantly throwing surprises even for the people studying it. The EU FP7 funded Project Getaway is currently looking at designing and trialling innovative and intelligent emergency signage systems for the evacuation of transport terminal passengers. As part of the recent work carried out by Professor Galea’s team at the Fire Safety Engineering Group, University of Greenwich, it was demonstrated that only 38% of people walking straight towards an emergency exit sign would actually see and register that it was there. And out of those 38%, not even all would decide to follow the sign (98%).
In Professor Galea’s words, ‘that indicates that signs can be effective if people see them – but they have to see them and register them. To assume that most will see them and react appropriately is the simple-minded approach to evac modelling. We’ve done a lot of research into the behaviour of people and we’ve found for example if you are right-handed you have a higher propensity to turn right – if you don’t know which way you are going. If you are left-handed, then left. And if you are right-handed and you come from a country that drives on the right then you have an even greater propensity to go right. So there are a variety of factors that makes this area quite complex.’
As part of Project Getaway the Fire Safety Engineering Group (FSEG) is looking at doing two things. Firstly, carrying out human behaviour experiments using intelligent dynamic signage – standard exit signs with flashing diodes dynamically pointing in the direction to follow, developed by Getaway partner Evaclite. ‘These diodes try to increase the affordance (or registered visibility) of the signs and we will repeat the experiments that gave us the 38% results and we are hoping we will get more than 38%.’
The second part involves the addition of ‘intelligence’ to the evacuation process. Getaway partner, Vision Semantics Limited (VSL: a spin off from Queen Mary University of London) is a leader in the Development of CCTV analysis, and it is developing an algorithm that can count people from CCTV footage and register where they are.
‘That information will provide the data for initial conditions in our evac software EXODUS. Running in faster-than-real time, EXODUS will then look at the different evacuation routes and take in information from heat and smoke sensors, and then calculate the best evacuation route. The special signs will then flash in the selected directions, whilst a flashing red cross will appear on signs to “kill” a particular direction.’
Professor Galea and his team are hoping that the addition of diodes to the signs will increase sign registration to 70% or more. ‘First of all if we can get more people following these types of signs then you may not necessarily need more intelligence – and if that works and nothing else then we would be over the moon. There is a lot of faith in signs but if nobody follows them…’
Professor Galea ascribes the low 38% ‘visibility’ factor to ‘learned irrelevance’, a phenomenon that in this context refers to people’s brains not registering an object that in their daily lives has become ‘irrelevant’ through over-familiarity and seldom – or ever – being used in anger. ‘That is why it is important that the flashing diodes only operate in an emergency because if you use them for normal way finding, people’s brains soon learn to ignore them.’
FSEG carries out a multitude of both large scale and lab-scale experiments where the subjects are not aware that they are in an experiment. A recent experiment (EU FP7-funded Project BeSeCu, Behaviour Security and Culture) by FSEG sought to understand the influence of culture on behaviour. Four unannounced trials evac trials took place in Czech Republic, Poland, Turkey and the UK, all involving the same types of buildings; population demographics; population activities; and fire alarms. ‘We found significant statistical differences in how quickly people responded to an alarm. We also looked at how response times were distributed, and we found each case was statistically different, suggesting that culture is an important factor.’
Furthermore, FSEG used the project to also collect definitive validation data sets. Validation data sets are crucial in order to test the performance of computer models, because they provide the real-life data.
‘One of the things that is missing from traditional validation data is where people actually started to evacuate from. The easy thing is to measure how long it took people to get out, but if you don’t know where they started from, then evac modellers can just continually relocate their agents (ie simulated pedestrians) until they get something that looks good and matches the validation data.’
In the case of Project BeSeCu, video cameras were placed throughout all four buildings to collect the location and behavioural data.
Professor Galea believes this is the first time that this type of data has been collected for complex multi-story buildings: ‘In the past modellers have been able to “be economical with the truth” when doing comparisons between validation data and simulations by running a number of simulations and waiting until they get a curve that looks like the validation curve. By nailing down the starting location and response times of trial participants we’ve reduced the wriggle factor.’
Developments – EXODUS
For the last 25 years’ Professor Galea and his team have been feeding their work into the EXODUS family of evac software and today the organisation has simulations for different environments: airEXODUS for aircraft evacuation, maritimeEXODUS for ship and off-shore evacuation, buildingEXODUS for evacuation in the built environment. In fact, FSEG was the first to incorporate the impact of fire into evac simulation.
FSEG was also the first to identify ‘exit hesitation’ time for aircraft evac, where evacuating passengers hesitate momentarily before committing to jump onto the slide. ‘If you used a building model and tried to model an aircraft you would get completely different type of behaviour. Those small hesitations make a big difference in the end. I do say that no one model can simulate all different environments as there are so many nuances to behaviours in special circumstances.’ Similarly, also represented in airEXODUS are evacuating passengers climbing over seats to circumvent the exit.
For the last four years FSEG has been working on the so-called ‘hybrid’ model of EXODUS, which in ‘tech speak’ combines the three traditional methods of descritising space in pedestrian and evacuation models. The ‘continuous’ approach takes measurements in a continuous way so any agent can occupy any point in space – this provides the greatest accuracy but also the highest processing/time costs especially when modelling 1000s of agents. The next approach is the ‘fine nodal approach’ where space is divided into regions and each agent can occupy one of those regions (the EXODUS approach). The third approach is the ‘coarse nodal’ approach which uses large portions of assigned space (eg entire rooms) and you only know that the agent is in the node (room) but you don’t know precisely where they are in the room.
A version of EXODUS has now been created which uses all three approaches so that modellers can mix and match. ‘The reason we’ve done this is sometimes you may want to have the added accuracy of the continuous approach in some regions of the geometry, with the bulk of the model for fine node, and for part of the geometry you may want to model crudely with large nodes. This also allows modelling of urban-sized geometries, and we are working on an EU FP7 project called IDIRA with this because we are modelling cross boarder disasters such as earthquakes and floods or even tsunamis.’
Another project has involved working to understand how people with disabilities exit structures with assistance equipment such as evacuation chairs, stretchers, carry chairs and drag sheets. ‘We did a series of experiments to characterise the performance efficiencies of these devices. So how long it would be to take someone 11 floors down using these devices. What would be the speed, how long they would need to stop and rest, and how others could overtake them.’ The paper will be presented at the Human Behaviour in Fire Symposium 2012 in Cambridge (September), and down the line, this collected data will be used to create accurate representations of these devices.
However, some preliminary simulations have been run using the data collected from the experiments to approximate the movement of combinations of device handlers, the device and the patient within EXODUS. Although the exact results are still to be released, Professor Galea commented that the time it took to evacuate a ward of 28 non-ambulant patients down 11 floors by a night staff of four female nurses, based on the EXODUS simulations was ‘shocking for some of the devices’, and he revealed that evacuation chair-type devices were by far the quickest according to the trials and the simulations. ‘Since we will have the capability to accurately represent these type of devices, we will be extending the application to airports and supermarkets with other types of devices – trolleys. Most evac models don’t incorporate an accurate representation of trolleys and the ability to be able to predict how long it takes the person with the device to turn a corner, for example, will be a powerful capability to have in these models – not just for evac but also pedestrian circulation.’
Evacuation from ships has been a large feature of the news in recent months and FSEG has done some work on evac from large cruise ships under the EU FP7 Safeguard Project. Whilst IMO (International Maritime Organisation) already stipulates how evac models should be used in evacuation analysis of passenger ship designs, dictating the appropriate scenarios and how they should be run, a number of weaknesses have been identified by FSEG and the Safeguard consortium. ‘Some of the data required in the ship evacuation analysis are the passengers response times, which are currently not based on data which is appropriate for all types of passenger ship. Also, there are no validation data sets for ship evacuation and so it is not possible to determine how good ship evac models are in predicting reality.
So we have carried out five semi-unannounced, at-sea experiments on three different classes of ship, and collected the response time data. In addition, we collected validation data, similar to the BeSeCu project for buildings, which will enable ship evac models to be tested and validated. Consequently we will be recommending to the IMO that in the future the more representative response time data generated in the Safeguard project should be used in design analysis and any evac models used in passenger ship design should be able to simulate that validation data set correctly. This will result in a kind of gateway into evac modelling for maritime environments, and those models that cannot do it won’t be used. The Safeguard consortium will also be suggesting a range of additional scenarios to test in a ship design.’
Fore more information about Project Getaway visit: http://fseg.gre.ac.uk/fire/GETAWAY.html
The FSEG youtube page has many animations of simulations generated using its software and several public lectures: http://www.youtube.com/FSEGresearch/