I wanna to estimate population in crowd environment. I am starter in this field, can anyone help me?
Take a look at dynamical systems. These deal with problems such as population estimation, by describing a system in an initial state (for example in terms of population) and then trying to understand what future states follow from changes in that initial state(Wikipedia). I remember using this to build a C program many years ago while studying at the university, to estimate the evolution of a rabbit population, in connection with the evolution of some predator population. One surprising thing was that if there were too few predators, the rabbits would start dying from starvation because they would become too many for the food available.
Good luck.
tnx for your response fernando,
your response is understanding future states but, I wanna enumerate people in sequences of image.
enumerate in crowd environment (not accurate and estimation is enough)
Dear Erfan
you can use many techniques for forecasting such as Artificial neural networks ANN or formulate the problem as an optimization problem where the decision variables are the coefficients of an ARIMA(X) model taking into account the population in the past years for example (Population(next year)=a0+a1*population(past year)+a2*population(two past years)
This may not be much help, but I think you will have to use some sort of statistical extrapolation method. Try to divide the image into maller sub-sections, and use a slow but more precise method of counting persons in the smaller sub-sections. Then extrapolate that to the larger image. So you might divide the image into an NxN grid (N^2 cells) and count people in a randomly chosen N of the cells, and multiply that count by N.
I don't know how to identify people; With the caveat that I have never done any image processing, I have seen transformations that turn images into basically line drawings. I would probably aim to recognize continuous outlines that match some percentage of a template for a person's head. Since some people are blocked by others, maybe 35%.
Then I would relate that computerized count to the actual known count (or count produced by a human being if the exact answer is not known) to find a statistical relationship to transform the computerized counts into the known counts, or a reasonable estimate of it.
Dear Sabhi,
I do not want to forecasting, I want to enumerate persons in a populate.
Dear Anthony,
can u give me references (paper, book, site, research ,....) for this (divide and enumerate and extrapolate method)? this way seems that is my solution.
for second section of your response, I say no. I do not want to recognize person and enumerate is enough.
Sorry Erfan, I have no particular paper to cite, this is just my common sense from many years of statistics.
I fail to see how you will "enumerate" people in an image without "recognizing" people in an image. That makes no sense.
Oh, I don't mean "recognize" them as individuals, I just meant "recognize" that some piece of an image *IS* a part of a person, not a car or podium or dog or tree!
I am not talking about recognizing the person in the crowd is a *particular* person, just that the image fits the profile of a human's head, like the gray-and-white avatar image next to my name.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms