Safer crowd evacuation from confined spaces
What if the unthinkable happens? You are in a confined space with tens of thousands of people during an event and you need to evacuate. Research and technologies are being developed at IT Innovation to help you get to safety.
In December 2015, IT Innovation went to the Anoeta stadium in San Sebastian, Spain, to run carefully designed scenario experiments to record crowd behaviour during evacuation. With a capacity of over 30,000 people, the Anoeta stadium hosts football matches, athletic sporting events and non-sporting events such as concerts.
Being able to evacuate people safely and quickly is of paramount importance to the security managers of the stadium. The safe evacuation of a crowd is challenging. It requires continuous knowledge of crowd behaviour in real-time and context information about the various spaces evacuees occupy. As result, IT Innovation is conducting research on the automated detection of crowd behaviour in confined spaces using computer vision, statistical mechanics concepts and knowledge models. This will revolutionise current crowd evacuation methods and strategies around the world since it will improve the management of crowds in real-time with advanced situation awareness and rapid response to critical events by security agencies.
IT Innovation led the development of the experiment together with partners in the eVACUATE project. This included the management and video recording of evacuation drills, which provided valuable source of information for further research to be conducted on crowd behaviour during evacuation. The available data on crowd in the public domain is usually incomplete and lack important information for labelling data prior to training computer vision tools. Notwithstanding the lack of information about the confined space in which the video scenes have been generated. Hence, the focus is on experimenting on good knowledge about the geometric details of the confined spaces, camera calibrations and context knowledge about how individuals or groups of individuals could potentially influence the spread of behaviour within the crowd for a given situation. The resulting labelled video dataset which we generated is being used to develop and test our new computer vision techniques for automatically detecting and interpreting groups’ motions and behaviours. The generated experimental dataset on crowd behaviour and their safer evacuation from confined spaces is of paramount importance for advancing the state of the art in human behaviour detection.
The eVACUATE project team, Drs Zoheir Sabeur, Banafshe Arbab-Zavar, Lee Middleton and Gianluca Correndo, who contributed in the development of the experiment said:
We carefully designed experiments to create and record complex crowd behaviour in different evacuation scenarios. It was essential that people participating in the experiment would behave as close to how they would in a real evacuation scenario. People received information about their default behaviour and we gave certain individuals a description of roles to play, to enact various scenarios, such as leaving an object down the exit space, moving against or even obstructing the flow of the crowd.
The technologies that is being developed at IT Innovation will help operational and security staff at venues such as the Anoeta football stadium make more advanced critical decisions about the safe evacuation of people. The machine detection of crowd behaviour with added value context information and reasoning modules will be integrated into a decision support system developed in the eVACUATE project, together with real-time optimisation on crowd evacuation routes.
We are focusing on detecting unusual behaviour, which can be flagged as alerts to the operational staff at the venue to help them more effectively and efficiently facilitate the evacuations. Zlatko Zlatev, project manager at IT Innovation for the eVACUATE project
The Anoeta stadium is one of four pilots in the eVACUATE project. The three other pilots include an international airport, cruise ship and a metro station. Our next development will be focused on validating our approach for detecting crowd behaviour with advanced context knowledge on all four confined spaces by the end of the project in April 2017.
To learn more, visit the eVACUATE project page.