CoherenceTheoryOfTruth on
open source crowd estimates:
InstaPundit talks about estimating crowd sizes. I’ve been to a couple of large rallies now; predictably the media lowballs (drastically, as in the case of a rally I went to in San Francisco in 1991) the organizers (probably) highball. I was thinking about the issue as we drove West on Friday. A simple, probably good enough, method would be to get 10 or 20 (or whoever is needed to cover the route with the needed granularity) people who know the march route and buildings lining it. They all climb to the top (or some high point) of a different building, a block (best), two or three apart. They then all click a picture (digital, 2 megapixal minimum, probably), at the same time, of their assigned area, presumably after the first marchers have reached the destination. After the event, take the pictures back to CountCentral (TM) where you divide the picture into 100 squares (or whatever is doable given the size of the scene). You count the number of people in 1 square then, looking around the rest of the image, estimate whether this square is more or less densely packed and adjust the figure to be “average.” Then multiply your estimate figure by 100 (or whatever amount you decide on above). Add the figures of all the pictures and you have an estimate.
But you’re still not done. Now you post the pictures and the methodology used on the net and invite comments (or improvements of the algorithm from someone who’s actually done stuff like this). In the best case some CS guy who’s into digital visual analysis will step up and offer to run the image through a program which estimates bodies based on their outline characteristics). Viola! “Open Source crowd estimates!”