[hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

[hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

Yves Quemener
Hello everyone,
I am looking for video recordings of big crowds (ideally protests) in
order to test various crowd-counting algorithms.

We had protests here in France recently and the differences between
police estimation and organizers estimation is unacceptably high (14 000
vs 120 000). Exploring the possibilities for a provable automated or
semi-automated count seems like a logical consequence for that. Both
sides are using approximations and multiply an estimated density by an
estimated area. Wouldn't it be possible to give a count of individuals ?
Various algorithms are promising for this (like this one :
http://mi.eng.cam.ac.uk/research/projects/Crowds/).

Therefore I would be very interested if someone could help me find
recordings of protests. It doesn't have to be free of rights, I am
willing to agree with restrictive licenses as I know that privacy can be
an issue on this kind of pictures (the license has to authorize research
use and open source software creation, though). I am interested in any
kind of video, but here is the ideal recording :
- recording of a protest where the crowd is at least 10 persons wide
- fixed view (not handheld)
- long enough so that most people in the first frame have passed by the
camera and are not visible anymore in the last frame
- Angle : either vertical enough so that one can see the ground
(vertical camera on a balcony) or facing the crowd (the typical picture
angle that journalists use to give a sense of the depth of the protest)
- high definition : the higher the resolution, the better. If the head
of a person is not at least a few pixels wide, it will be hard to count.
- Without too much compression artifacts (either with a good quality
compression or uncompressed)

This list of criterions is almost impossible to fulfill I fear, so I am
interested in any video that fits at least one criterion. I will try to
make some recordings on rush hour in strategic places, but a protest
moves differently than a rush-hour crowd. Comments and advices are of
course more than welcomed. And if someone heard about a project with
similar goals, please let me know !

Iv
_______________________________________________
Discuss mailing list
[hidden email]
http://lists.hackerspaces.org/mailman/listinfo/discuss
Reply | Threaded
Open this post in threaded view
|

Re: [hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

pieter.heremans
hey
maybe it's possible to test your algorithms on video captures from
sports events i.e. cheering public at soccer game or the start of a
marathon or something ?  i guess these data will be easier to find and
a lot less privacy-sensitive.
i found this one: 'crowd_run' on http://media.xiph.org/video/derf/

ptr_
(hsbxl - hs brussels,belgium)




On Wed, Jun 30, 2010 at 2:03 PM, Yves Quemener <[hidden email]> wrote:

> Hello everyone,
> I am looking for video recordings of big crowds (ideally protests) in order
> to test various crowd-counting algorithms.
>
> We had protests here in France recently and the differences between police
> estimation and organizers estimation is unacceptably high (14 000 vs 120
> 000). Exploring the possibilities for a provable automated or semi-automated
> count seems like a logical consequence for that. Both sides are using
> approximations and multiply an estimated density by an estimated area.
> Wouldn't it be possible to give a count of individuals ? Various algorithms
> are promising for this (like this one :
> http://mi.eng.cam.ac.uk/research/projects/Crowds/).
>
> Therefore I would be very interested if someone could help me find
> recordings of protests. It doesn't have to be free of rights, I am willing
> to agree with restrictive licenses as I know that privacy can be an issue on
> this kind of pictures (the license has to authorize research use and open
> source software creation, though). I am interested in any kind of video, but
> here is the ideal recording :
> - recording of a protest where the crowd is at least 10 persons wide
> - fixed view (not handheld)
> - long enough so that most people in the first frame have passed by the
> camera and are not visible anymore in the last frame
> - Angle : either vertical enough so that one can see the ground (vertical
> camera on a balcony) or facing the crowd (the typical picture angle that
> journalists use to give a sense of the depth of the protest)
> - high definition : the higher the resolution, the better. If the head of a
> person is not at least a few pixels wide, it will be hard to count.
> - Without too much compression artifacts (either with a good quality
> compression or uncompressed)
>
> This list of criterions is almost impossible to fulfill I fear, so I am
> interested in any video that fits at least one criterion. I will try to make
> some recordings on rush hour in strategic places, but a protest moves
> differently than a rush-hour crowd. Comments and advices are of course more
> than welcomed. And if someone heard about a project with similar goals,
> please let me know !
>
> Iv
> _______________________________________________
> Discuss mailing list
> [hidden email]
> http://lists.hackerspaces.org/mailman/listinfo/discuss
>



--
http://www.L45.be/voidpointer
0493 52 5009
_______________________________________________
Discuss mailing list
[hidden email]
http://lists.hackerspaces.org/mailman/listinfo/discuss
Reply | Threaded
Open this post in threaded view
|

Re: [hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

Yves Quemener
On 06/30/2010 02:38 PM, pieter.heremans wrote:
> hey
> maybe it's possible to test your algorithms on video captures from
> sports events i.e. cheering public at soccer game or the start of a
> marathon or something ?  i guess these data will be easier to find and
> a lot less privacy-sensitive.
> i found this one: 'crowd_run' on http://media.xiph.org/video/derf/
>    
That's an excellent idea ! Plus it is possible to find high quality
video of these. Cheering public is trickier I think because individuals
are not moving independentlyand are usually facing in a different
direction than the apparent movement.
_______________________________________________
Discuss mailing list
[hidden email]
http://lists.hackerspaces.org/mailman/listinfo/discuss
Reply | Threaded
Open this post in threaded view
|

Re: [hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

Steve Clement
Flash mob Videos are quite good for "larger" crowds.

cheers,


Steve
--
Let's build a Hackerspace in Luxembourg!
http://www.hackerspace.lu
mailto:[hidden email]
.lu: +352 20 333 55 65

On Jun 30, 2010, at 3:01 PM, Yves Quemener wrote:

> That's an excellent idea ! Plus it is possible to find high quality video of these. Cheering public is trickier I think because individuals are not moving independentlyand are usually facing in a different direction than the apparent movement.

_______________________________________________
Discuss mailing list
[hidden email]
http://lists.hackerspaces.org/mailman/listinfo/discuss
Reply | Threaded
Open this post in threaded view
|

Re: [hackerspaces] Looking for video recording of protests/crowds for testing crowd-counting algorithms

Yves Quemener
On 07/01/2010 09:59 AM, Steve Clement wrote:
> Flash mob Videos are quite good for "larger" crowds.
>
> cheers,
>
>
> Steve
>    
Sounds good but it is hard to find good quality movies of these. I'll
try to find one of a crowd not dancing synchronously (oh why ?) and of
good quality. If you have links do not hesitate to post them. The good
thing about these is that they can be long and from a fixed point of view.
_______________________________________________
Discuss mailing list
[hidden email]
http://lists.hackerspaces.org/mailman/listinfo/discuss