Visual surveillance of crowds is an important part of event management as well as policing. Now, a team from Malaysia and Saudi Arabia have looked at the various tools that have become available in recent years for automatically assessing the number of people in a crowd and determining the dynamics and movement of that crowd. Writing in the International Journal of Computational Vision and Robotics, the team finds several gaps in the current state-of-the-art technology and points developers to how those gaps might be filled.
Huma Chaudhry and Mohd Shafry Mohd Rahim of Universiti Teknologi Malaysia, Tanzila Saba of Prince Sultan University, Riyadh, and Amjad Rehman of Al Yamamah University, also in Riyadh, point out that computer vision research has moved towards crowd control and management in recent years with a view to addressing issues of security and safety when large numbers of people are gathered in one place. The fundamental problem that has to be addressed is how to manage multiple data streams from closed-circuit television (CCTV) and other sources that monitor crowd dynamics at events, in busy towns and cities and elsewhere. There a limit to how visual assessment of CCTV and so automated, computerised solutions are needed.
The team highlights some major events where there have been numerous casualties. Sometimes casualties at some events might be fewer than 100 people, but larger events might see thousands of casualties over a prolonged period. Automated crowd assessment could open up new ways t understand crowd dynamics and reduce those numbers. Some of the same insights from aerial crowd surveillance and other methods might also help in disaster relief activities where large numbers of people might be present in a given location.
Chaudhry, H., Rahim, M.S.M., Saba, T. and Rehman, A. (2019) ‘Crowd detection and counting using a static and dynamic platform: state of the art’, Int. J. Computational Vision and Robotics, Vol. 9, No. 3, pp.228–259.