images to video - Object tracking

[Anusha]

Dear bolme ,

I want to display all the log images as video . 

how can I do that ?

or 

I want to track object in real time , I have modified code using mosse filter . 

here are some doubts .

1) how to re select object if it is out of frame ?

2) how to track multiple objects?

3) Is there any other ways to detect object in first frame ? / I dont want to specify object in first frame , it has to detect by using few first frames. 

thanks and regards .

yogesh

[David Bolme]

For creating a video from an image I typically use ffmpeg.  The command will probably look something like this but you will have to read the documentation and make sure the filenames are correct.

$ ffmpeg -r 30 -i %06d_FRAME.jpg out.mov 

For your other questions.  The MOSSE tracker is only designed to track and object that has either been selected by the user or detected by a different algorithm.  It is possible to track multiple objects at the same time by creating multiple tracker objects but you will need to call the update method on all of them for each frame but it should be fast enough that you can track many objects simultaneously.  Once the object goes out of frame it can be very difficult to redirect.  You will need to at provide an update location when it reenters the frame but if it has changed appearance you will probably have to restart the tracker.

You could try to create a MOSSE filter that can detect the objects you are interested in.  Some you can correlate a new frame with the filter and apply a simple threshold to detect objects, but if there are scale and rotation changes to your targets MOSSE is probably not a good solution.  You can use the MOSSE Trackers to follow objects that were detected using other methods like Viola and Jones or Deformable Part Models.

Good Luck,

Dave

--

---
You received this message because you are subscribed to the Google Groups "pyvision" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pyvision+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[yogesh bn]

Thank you very much sir

--

---
You received this message because you are subscribed to a topic in the Google Groups "pyvision" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/pyvision/fYWKS2w7uRs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to pyvision+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

YOGESH BN
ME-SSA
EE dept.
IISc bengaluru

yogeshbn@ee.iisc.ernet.in

9964106764
9164610640

__________________________

" WINNERS DON'T DO DIFFERENT THINGS

THEY DO THINGS DIFFERENTLY "

[yogesh bn]

Dear bolme ,

May I know what are the different 9 transforms you have used in MOSSE filter .

And is there any particular reason to use that affines ?

On Tue, May 27, 2014 at 1:53 AM, David Bolme <dbo...@gmail.com> wrote:

--

---
You received this message because you are subscribed to a topic in the Google Groups "pyvision" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/pyvision/fYWKS2w7uRs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to pyvision+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[David Bolme]

I would suggest original image and all combinations of +/- 10% scale and +/- 10 deg of angle. That should give you 9 transforms that cover what you expect to see in the next few frames. See my code if you want exact details. 

--

David Bolme 

Sent from my phone.