Nel 2008/2011 ho lavorato attivamente su rilevamento automatico di oggetti in immagini, definendo un metodo per velocizzarlo ed aumentarne l’accuratezza. Il metodo si basa su tecniche di sampling Monte Carlo ed è stato recentemente pubblicato su IEEE Transactions on Pattern Analysis and Machine Intellingence

G. Gualdi, A. Prati, R. Cucchiara, “Multi-Stage Particle Windows for Fast and Accurate Object Detection” (to be published soon) IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012

Riporto qui di seguito alcune immagini di rilevamento di pedoni e volti e un video di esempio:

cmuB-t.jpg

cwsiB-t.jpg

graz02B-t.jpg

inriaB-t.jpg

12 Responses to Rilevamento di Oggetti

  1. Ajay George says:

    Sir,
    I am trying to implement your paper as a project. I wish I could get your help on that.

  2. lv fenghua says:

    Dear Professor,
    Could you tell me how to download the code of your method? I’m focus on the field of object detection! I expect for your reply!
    Wish you have good health and good luck!

    • Giovanni says:

      Dear Fenghua,

      it is not our intention right now to publish the code, so the best way is for you to reimplement that; if you understand how it works, it shouldn’t take long; don’t hesitate if anything in the method is unclear,

      best regards,
      Giovanni

  3. kyo Wu says:

    Dear Dr. Gualdi,

    I’ve read your paper “Multi-Stage Particle Windows for Fast and Accurate Object Detection”, and I’m interested in the CWSi and CWSv datasets.

    But I couldn’t find the download link at http://imagelab.ing.unimore.it/visor.

    Could you tell me how to download the datasets?
    Thank you

    • Giovanni says:

      Hello Kyo,

      please go to http://www.openvisor.org/ then, you need to log-in, since the dataset is visible only if you are logged in. As soon as you log in, you will see “Contruction working sites (8 items)”, i.e. 1 for CWSi and 7 for CWSv.

      I have requested to the OpenVisor administrator to make the dataset public, however it may take a few days.

      let me know if you still have problems
      Gio

  4. Trung Quy says:

    Dear Dr Gualdi,

    I am reading your paper, I don’t understand how can you apply your method to non-cascade classifier (such as SVM with HOG).You show the experimental results with SVM but I don’t understand how to implement it. Could you explain in detail about it?

    If necessary, we can discuss more via email.

    Thanks.

    Trung Quy.

    • Trung Quy says:

      What is “stage” in term of non-cascade classifier?
      Thanks

    • Giovanni says:

      Dear Trung,

      sorry for the late reply…
      here the reply:
      the SVM returns a margin, that is the distance from the boundary: you can use that as the value leading the particle scattering.
      Since we want to work over [0..1] and the SVM margin is usually (-inf… +inf), we use a sigmoid function: see eq 3 of the PAMI.

      best regards
      Giovanni

  5. qiuzhe says:

    Dear Professor,

    Recently,I’ve read your paper “Multi-Stage Particle Windows for Fast and Accurate Object Detection”.I’m confused about the responses of the SVM. In this paper,you said “i.e., the distance to the class-dividing boundary”.could you explain it in detail?
    Thank you very much!

    • Giovanni says:

      Hi, SVM for binary classification provides a margin, i.e. a real number that is the distance from the best iperplane dividing one class from the other. Please refer to documentation on binary classification with SVM for further details;
      b.r.
      Giovanni

      • qiuzhe says:

        Dear Professor,

        Thank you for your reply!it seems that in the algorithm 1 of the paper “Multi-Stage Particle Windows for Fast and Accurate Object Detection”,you get a set of particle windows to which the response of the classifier is high and finally you use these windows to construct a likelihood function.is that true? so how do we get the final bounding box? sample it from the likelihood function?
        I’m looking forward to your reply.Thank you very much!

Lascia un Commento

L'indirizzo email non verrà pubblicato. I campi obbligatori sono contrassegnati *

*

È possibile utilizzare questi tag ed attributi XHTML: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>