It’s always impressed me whenever I’ve watched big sporting events like the Tour de France, how the commentators have been able to confidently quote the number of people stood by the road watching the riders zoom by.

Apparently, these estimates are typically done by scanning multiple aerial photographs manually, with the number of heads per inch counted and added together.  A new project from researchers at the University of Central Florida claims to have automated the process for the first time.

Automated crowd counting

The traditional approach to doing this would usually take up to a week to complete, but the researchers, from UCF’s Center for Research in Computer Vision, claim to be able to complete it accurately in around half an hour.

They believe that this speed will give event organizers access to critical information for better event planning or even emergency response.

The researchers put their system through its paces during a test run in September, when thousands of demonstrators descended on Barcelona to call for independence of Catalonia.

The software trawled through 67 aerial images of the gathering, with a person count for each image generated in a matter of minutes, and a final estimate of 530,000 produced.

The final count was then compared with official figures produced by the organizers of the rally, who had released numbers significantly higher than those given by the UCF team.

“Automated computer analysis of such large-scale and dense crowds has never been done before,” the team say. “We will continue to push the envelope of state-of-the-art in-crowd analysis so that it can by help the authorities and governments manage real-time safety of large crowds and perform post-event analysis of such gatherings.”

The project has built upon previous attempts to automate the counting of crowds.  Those early efforts were typically done on small’ish groups, so performing the task on a crowd of 500,000 people is a significant upping of the ante.

“This is a huge milestone in how mass crowds are counted,” the team proudly declared. “UCF’s computer-vision program has been the leading research group on developing computer-vision techniques for analysis of crowd videos and their research can have significant impact in terms of development of next-generation crowd-management technologies.”