How to measure the number of people in a picture of a crowd?

Question

Background: Israel (and the middle east in general) is filled with protests.

I am curious, when given a picture, to estimate how many people are in that picture (often a picture of a large crowd).

What modelling can offer some solution to this problem? (and, of course, can it be done with any open source package. Say, R?)

possible duplicate of How to estimate how many people attended an event (say, a political rally)?
Hi cardinal - thank you for the link. The question I ask is more specialized to a specific case (which is not answered in the link you give). I am not sure there is an answer, but still. With regards, Tal
Is it a picture of the whole crowd? Is the picture taken from above or is it taken at eye level? To have success you'll want a standard protocol for what counts as a picture.

user12287 · Accepted Answer · 2012-07-26 17:08:18Z

I am working in a similar project, I will follow the next approach:

Get a lot of classified images, images with few people, images with crowded spaces. For example, 'The Zócalo' in Mexico, City could afford more or less 250 000 people.
Extract features from these images, maybe with methods like either HOG or SIFT. HOG for example, is widely used to extract features in projects aimed to detecting pedestrians. Visit http://hogprocessing.altervista.org/
With the data obtained in the above step, is possible run some machine learning algorithm: SVM or NN. It will be necessary train this algorithm, and thereafter when you have a new image, then is possible use the NN or SVM trained in order to get a prediction.

I guess you can follow a similar path.

1 Answer