Re: Empirical crowd estimates

From: spike66 (spike66@attbi.com)
Date: Sat Feb 22 2003 - 14:02:17 MST

  • Next message: Wei Dai: "Re: IRAQ: cost of inspections"

    Anders Sandberg wrote:
    >
    > Hmm, do you mean how much people arrive and depart? One might make an
    > estimate by looking at two subsequent photos and matching people (using
    > a color similarity + distance metric, and a fast bipartite graph optimum
    > matching). That would produce an estimated movement field...
    > It is somewhat related to the algorithms used in ant tracking I sent a
    > few weeks back to Spike, although this does apparently scale better.

    Yes and that was way cool. I am still thinking about
    ant metrics, but this crowd density thing might
    be even more interesting. Looks to me like we could
    estimate the density of a crowd based on the way it
    reflects radar. We know that radar can be cast over
    wide areas. I propose a test where we get about a
    thousand volunteers, have them stand on an area the
    size of a football field, take a radar image, then
    get them to stand between the 20s, take another image,
    bring them inside the 40s, at which time it would be
    about as densely packed as any crowd ever gets, take
    an image.

    Then from that, we could compare radar images of a
    rally, create a crowd density contour map with
    iso-humes, or lines connecting regions of constant
    human density and from that estimate how many humans
    are present. One such radar image per minute and
    one could estimate the number of human-hours. Way
    cool Anders! spike



    This archive was generated by hypermail 2.1.5 : Sat Feb 22 2003 - 14:04:37 MST