Headcount without helicopters
December 8, 2009 1:43 AM
How do you accurately estimate the size of a crowd?
Mrs aqsakal and I went to a demo in Rome on Saturday. We (and n thousand others) walked 2.6 km. from the starting point to the end point, where there was a stage and a megascreen and music and speeches. The police told the media there were 90K of us. I later read on the BBC website and others that the organisers had told the media we were 350K.
Of course the government was interested in minimising the attendance, and the organisers in maximising it. This "war of statistics" has happened at every demo I have ever attended. What I'd like to know is, how does one go about making an accurate, impartial estimate?
Some reports, confirmed by friends who also took part, said that when the head of the procession had already reached the destination, there were still people arriving at the starting point and heading off down the route. So, if these reports are correct, you had a two-and-a-half km. column of people, of varying density, walking along roads of a varying but roughly similar width.
Mostly it was very crowded, physical contact unavoidable, but occasionally there was a gap where some had moved ahead faster and others behind them had walked at a more leisurely pace, set by the truck with the amp and the music. Occasionally there were side-tracks, where people (like us) who get angst attacks in thick crowds, and who knew the terrain, took a quieter side route and rejoined the main column somewhere else. But in general this was a cheerful, non-violent demo with a fairly homogenous makeup, not a running battle between die-hards and cops like the Genoa G8 riots, or Athens on Sunday, so crowd estimations should have been fairly easy. Especially from the obligatory police chopper which was constantly circling overhead. But a difference on a 4x scale, even allowing for political bias - WTF?
I imagine one answer would be to get a high shot, vertical if possible, overlay a grid of an appropriate cell size, count heads in a cell and multiply up. Except that chartering a helicopter is expensive, and access to strangers' rooftops not easy to obtain; all I have is my own crowd-level pictures. A really very, very rough "calculation" from my own pix makes the 350K figure look fairly realistic, but with all the caveats mentioned.
Can anybody offer some practical expertise on this?
Mrs aqsakal and I went to a demo in Rome on Saturday. We (and n thousand others) walked 2.6 km. from the starting point to the end point, where there was a stage and a megascreen and music and speeches. The police told the media there were 90K of us. I later read on the BBC website and others that the organisers had told the media we were 350K.
Of course the government was interested in minimising the attendance, and the organisers in maximising it. This "war of statistics" has happened at every demo I have ever attended. What I'd like to know is, how does one go about making an accurate, impartial estimate?
Some reports, confirmed by friends who also took part, said that when the head of the procession had already reached the destination, there were still people arriving at the starting point and heading off down the route. So, if these reports are correct, you had a two-and-a-half km. column of people, of varying density, walking along roads of a varying but roughly similar width.
Mostly it was very crowded, physical contact unavoidable, but occasionally there was a gap where some had moved ahead faster and others behind them had walked at a more leisurely pace, set by the truck with the amp and the music. Occasionally there were side-tracks, where people (like us) who get angst attacks in thick crowds, and who knew the terrain, took a quieter side route and rejoined the main column somewhere else. But in general this was a cheerful, non-violent demo with a fairly homogenous makeup, not a running battle between die-hards and cops like the Genoa G8 riots, or Athens on Sunday, so crowd estimations should have been fairly easy. Especially from the obligatory police chopper which was constantly circling overhead. But a difference on a 4x scale, even allowing for political bias - WTF?
I imagine one answer would be to get a high shot, vertical if possible, overlay a grid of an appropriate cell size, count heads in a cell and multiply up. Except that chartering a helicopter is expensive, and access to strangers' rooftops not easy to obtain; all I have is my own crowd-level pictures. A really very, very rough "calculation" from my own pix makes the 350K figure look fairly realistic, but with all the caveats mentioned.
Can anybody offer some practical expertise on this?
or, if you want to go the science route, this powerpoint set - the last slide has a lot of references.
posted by DreamerFi at 4:12 AM on December 8, 2009
posted by DreamerFi at 4:12 AM on December 8, 2009
Some folks at Boston University refined some uses of remote sensing (aerial photography and satellite imagery) in this context, in response to the crowd size estimating controversies surrounding the Million Man March. (Some interesting stuff along these lines linked in the footnotes of the Wikipedia page.)
posted by jjjjjjjijjjjjjj at 4:48 AM on December 8, 2009
posted by jjjjjjjijjjjjjj at 4:48 AM on December 8, 2009
In computer vision there's an active subdiscipline in crowd monitoring and modelling. As far as I know, none of these systems are in production yet, but they're going to be coming soon.
A good place to look for an overview is the PETS (Performance Evaluation of Tracking and Surveillance) 2009 workshop - you can find the datasets, proceedings and so on here:
http://www.cvg.rdg.ac.uk/PETS2009/
They had special challenges for person-counting, as well as person-tracking and behaviour monitoring. The short version is that it's very hard for a computer to count people from a digital video when it can't see each individual person separately, but they're getting there.
posted by handee at 5:03 AM on December 8, 2009
A good place to look for an overview is the PETS (Performance Evaluation of Tracking and Surveillance) 2009 workshop - you can find the datasets, proceedings and so on here:
http://www.cvg.rdg.ac.uk/PETS2009/
They had special challenges for person-counting, as well as person-tracking and behaviour monitoring. The short version is that it's very hard for a computer to count people from a digital video when it can't see each individual person separately, but they're getting there.
posted by handee at 5:03 AM on December 8, 2009
It is not often the case that the people with adequate resources to do an accurate count have any interest in reporting an accurate count.
posted by rikschell at 7:38 AM on December 8, 2009
posted by rikschell at 7:38 AM on December 8, 2009
The three methods I have used/seen people use
1. Count - stand at one point and count the people going by. It works up until a point. Even better if you have a partner and you average your results.
2. Guesstimate - When looking at a crowd pick out a group of people, anywhere from 10 to 100 ish. Take a mental picture of the area they are covering and then impose the perimeter of that area on the rest of the group and count how many times you can overlay it in your mind. Does that make any sense?
3. Communist - The commies I used to work with would just take whatever the cops said and double or triple it depending on their needs.
posted by nestor_makhno at 8:11 AM on December 8, 2009
1. Count - stand at one point and count the people going by. It works up until a point. Even better if you have a partner and you average your results.
2. Guesstimate - When looking at a crowd pick out a group of people, anywhere from 10 to 100 ish. Take a mental picture of the area they are covering and then impose the perimeter of that area on the rest of the group and count how many times you can overlay it in your mind. Does that make any sense?
3. Communist - The commies I used to work with would just take whatever the cops said and double or triple it depending on their needs.
posted by nestor_makhno at 8:11 AM on December 8, 2009
Many thanks to all. Some fascinating stuff there. The scientific part looks promising, but beyond most people's means. Counting legs or eyeballs and dividing by two seems the most accurate so far... I'm encouraged to see some experts agree with my amateur idea of counting a selection (either on the ground or on a pic) and multiplying up.
I take Fiasco da Gama's point of not writing police estimates off immediately as lies, but this implies a certain level of trust in police evidence which, at least in my country, is sadly lacking (see also the long discussion on the Amanda Knox/Mez Kercher trial on the blue).
Equally nestor_makhno's point 3 - that confirms my experience, too.
Thanks again, guys and girls.
posted by aqsakal at 5:52 AM on December 9, 2009
I take Fiasco da Gama's point of not writing police estimates off immediately as lies, but this implies a certain level of trust in police evidence which, at least in my country, is sadly lacking (see also the long discussion on the Amanda Knox/Mez Kercher trial on the blue).
Equally nestor_makhno's point 3 - that confirms my experience, too.
Thanks again, guys and girls.
posted by aqsakal at 5:52 AM on December 9, 2009
This thread is closed to new comments.
It's done, with moving demonstrations, by picking a point, preferably a narrow geographic choke point everyone has to pass, and by regularly estimating crowd depth through an arbitrary line, estimating the speed---a big rally rarely moves at more than 2-3km/h---and working out volume that way. The helicopter isn't there to estimate distance, as it's very difficult to work out density from above, it's there to provide traffic control for the surrounding streets.
If it's in a location where everyone has to use certain train stations, it gets even easier, as rail lines are able to estimate the volume of crowds and queues very accurately.
The police estimates oughtn't to be written off immediately as lies, BTW. At least in my country, the figure cited in the news will be the one also used by the police HR to prioritise staff, so they're liable to be questioned come Budget time should they grossly over- or under-estimate.
posted by Fiasco da Gama at 1:58 AM on December 8, 2009