I'm trying to do people detection from bird eye view.

Actually, it is not a normal bird eye view because I don't have RGB images but I have disparity map, depth map as well.

In these kind of images, people are shown like lines that represent a single side of the person. I try to re-explain it: these lines represent the front or the back of the person.

I searched for papers on internet but I got no results.

Anyone have any information about that?

PS: I'm working on Matlab.

The following image shows how people looking like from a bird-eye view depth map. You don't look at the big stain. The 3 red circles contain the bird-eye view depth map of (from the left to right): a street fountain, a person and a roadsign.

enter image description here

  • 1
    $\begingroup$ Show the images please. $\endgroup$
    – Maurits
    Oct 23 '12 at 23:14
  • 2
    $\begingroup$ Do you have multiple frames? What resolution are the images - the image you attached looks very lo-res! $\endgroup$ Oct 24 '12 at 14:04
  • $\begingroup$ That image is used only to let better understand my question. Those elements inside red circles are what I have to classify, actually, I have to distinguish people from other stuff. Anyway, I have 640x480 frames. $\endgroup$ Oct 25 '12 at 15:17

It appears that motion is your only cue in this case, assuming that you have a time-series of these images. I would start by trying to track these objects using a Kalman filter with a constant velocity motion model. The fountain and the street sign would be stationary. If your images contain other moving objects, e. g. cars, then you can try using speed to distinguish them from people.

Here is an example of how to track multiple objects using the Computer Vision System Toolbox for MATLAB. The example detects moving foreground objects using the Gaussian mixture models. You may want to replace that part or augment it using the depth information.

  • $\begingroup$ thank you Dima. That was my first thought. But, in some cases, people move in non-linear way(e.g. a person turns around and re-walk his step). In these cases, Kalman filter doesn't work properly. Do you have any other suggestion? $\endgroup$ Oct 29 '12 at 11:21
  • $\begingroup$ @user1768064, True, but most of the time people go somewhere, i. e. more or less in a straight line. So even if you cat track an object moving predictably for some amount of time, that may be enough to identify it as a person. $\endgroup$
    – Dima
    Oct 29 '12 at 13:51
  • $\begingroup$ @user1768064 My point is that I would try the Kalman filter first. If that doesn't work for you, then try more complex algorithms that can better handle non-linearity: extended Kalman filter, unscented Kalman filter, or particle filter. $\endgroup$
    – Dima
    Oct 29 '12 at 13:53
  • $\begingroup$ I will try that way! thank you very much Dima for the support. $\endgroup$ Oct 29 '12 at 15:57
  • $\begingroup$ I forgot to mention that in my scenario there could be other non-person moving around and I don't want to track them, I want to track only person. These things that could be into the scene are whatever that a person could take with him/her into a shop (i.e. a stroller)... Any help for how to detect person and non-person in this kind of scenario? $\endgroup$ Nov 7 '12 at 8:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.