0

I've read a lot about object/tracking detection. There are a lot algorithms in moving objects tracking such as frame differencing, background subtraction etc. Also there are a lot object detection algorithms: Viola Jones, Haar-like features etc. And I use Viola Jones algorithm, Haar-like features and cascade of classifiers to detect objects. So if I use classifiers for face detection for example, my question is how to track all faces on each frames.

Jav_Rock
  • 22,059
  • 20
  • 123
  • 164
Frankie Drake
  • 1,338
  • 9
  • 24
  • 40
  • 1
    Are Bayes filters (particle filter, Kalman filter, ...) what you are looking for? – alfa May 01 '12 at 07:38
  • Hmm, thank you.So do you want to say that any frame differencing methods are not useful here? – Frankie Drake May 01 '12 at 08:01
  • 1
    No, I just wanted to suggest some more methods. Bayes filters are better for noisy and partially observable environments (what happens when you cannot recognize the face in every frame, when it is hidden, etc.?). – alfa May 01 '12 at 08:24
  • Actually I need as easy as possible algorithm. However I would like to hear your advices anyway. – Frankie Drake May 01 '12 at 10:01
  • 1
    I don't know what would be the easiest possible algorithm for your use case. Kalman filters are a standard approach that is used e. g. for robot localization, state estimation, sensor fusion... You can find lots of papers etc. about that with Google. – alfa May 01 '12 at 12:25

1 Answers1

1

Simplest method is to just run the detector for every frame and see the detection overlaps to decide if its same face in next frame. or a sime euclidean distance might be good. This is good if face is always. If the face can be non frontal then you need to do some kind of histgram based tracking which is computationally inexpensive.

Krish
  • 1,747
  • 14
  • 19