Sepp

I am trying to create a patch, which analysis live video. The idea is to detect moving human shapes - which can enter the stage from any direction. Every shape has a different effect applied over them and as they encounter each other the effects should be interchanged between them.

With cv.jit.label I am able to detect shapes, isolate them and apply different effects to them...But I have huge problem with continuity. As you probably know cv.jit.label labels blobs uniquely, but when their order changes - i.e. which is top-left most or largest blob - blob loses its original label (1 becomes 2 etc.). This is of course problem as the idea is to switch effects between shapes.

Does anyone have a idea how to solve this problem ? Any help would be greatly appriciated.

I am trying to create a patch, which analysis live video. The idea is to detect moving human shapes - which can enter the stage from any direction. Every shape has a different effect applied over them and as they encounter each other the effects should be interchanged between them. 

cv-jit-label-question

As I write this email, I am working on a set of externals to do just that.

Essentially you have to start "tracking" the blobs based on various metrics

such as bounding-box overlap from frame to frame, color histogram from frame

to frame, and any number of other metrics. Then you have to find the

correspondences from frame to fram and relabel (renumber) the output from

The externals are not finished yet, but when they are, I'll let you know.

Aside from externals, one trick I have used (depending on the situation) is

to send a rotated binary matrix into cv.jit.label. This allows you to deal

with some situations where it would be better to let cv.jit.label count

vertically from the lower-left-hand corner rather than horizontally from the

As I write this email, I am working on a set of externals to do just that.
Essentially you have to start "tracking" the blobs based on various metrics
such as bounding-box overlap from frame to frame, color histogram from frame
to frame, and any number of other metrics.  Then you have to find the
correspondences from frame to fram and relabel (renumber) the output from
cv.jit.label.

Aside from externals, one trick I have used (depending on the situation) is
to send a rotated binary matrix into cv.jit.label.  This allows you to deal
with some situations where it would be better to let cv.jit.label count
vertically from the lower-left-hand corner rather than horizontally from the
upper left hand corner.

My 2up.jit.centroids object, available at 2uptech.com, may be of help 

to you if you are using a Mac. The help patch of the object 

demonstrates the behavior you are looking for.

My 2up.jit.centroids object, available at 2uptech.com, may be of help  
to you if you are using a Mac.  The help patch of the object  
demonstrates the behavior you are looking for.

---------------------------------------------------

-----------------------------------------------------------

Thank you for the advice. That matrix inversion tip might prove to be useful. Hopefully you succeed on writing those externals. I am certain that there are a lot users that require something like that. Can you predict when they could be ready?

----------------------------------------------------

My 2up.jit.centroids object, available at 2uptech.com, may be of help

to you if you are using a Mac. The help patch of the object

--------------------------------------------

Unfortunately I am not using mac. Have you planned on converting it to Win?

---------------------------------------------------
As I write this email, I am working on a set of externals to do just that.
Essentially you have to start "tracking" the blobs based on various metrics
such as bounding-box overlap from frame to frame, color histogram from frame
to frame, and any number of other metrics. Then you have to find the
correspondences from frame to fram and relabel (renumber) the output from
cv.jit.label.

Aside from externals, one trick I have used (depending on the situation) is
to send a rotated binary matrix into cv.jit.label. This allows you to deal
with some situations where it would be better to let cv.jit.label count
vertically from the lower-left-hand corner rather than horizontally from the
upper left hand corner.

Cheers,
Christopher
-----------------------------------------------------------

Thank you for the advice. That matrix inversion tip might prove to be useful. Hopefully you succeed on writing those externals. I am certain that there are a lot users that require something like that. Can you predict when they could be ready? 

----------------------------------------------------
My 2up.jit.centroids object, available at 2uptech.com, may be of help
to you if you are using a Mac. The help patch of the object
demonstrates the behavior you are looking for.

-Randy
--------------------------------------------
Unfortunately I am not using mac. Have you planned on converting it to Win? 

Quote: randall jones wrote on Sun, 29 January 2006 12:58

> I don't have a Windows machine here. If someone is interested in 

> compiling a Windows version, I can make the source code available.

Well, I would be interested. I have never converted mac externals to win though, but I could certainly try.

Quote: randall jones wrote on Sun, 29 January 2006 12:58
----------------------------------------------------

> I don't have a Windows machine here.  If someone is interested in  
> compiling a Windows version, I can make the source code available.
>
> -Randy
>
>
>
>
----------------------------------------------------

I'll be able to compile the externnals later tonight. It shouldn't

take me but a few minutes to do. Let me know and I'll post them ASAP.

Hi Randy,
I'll be able to compile the externnals later tonight.  It shouldn't
take me but a few minutes to do.  Let me know and I'll post them ASAP.

You know, I'm not sure when it will be working. My goals is to have 

I have positive preliminary results with test data (i.e. one blob per 

object with no merging or splitting of blobs ... a sort of closed 

world scenario), but it becomes more complicated when one is required 

to intelligently group broken blobs with holes, etc.

You know, I'm not sure when it will be working.  My goals is to have  
it working within the next month.

I have positive preliminary results with test data (i.e. one blob per  
object with no merging or splitting of blobs ... a sort of closed  
world scenario), but it becomes more complicated when one is required  
to  intelligently group broken blobs with holes, etc.

OK, Wesley Smith has compiled a 2up.jit.centroids external for 

OK, Wesley Smith has compiled a 2up.jit.centroids external for  
Windows.  Thanks Wes!  Big up community!

Thanks a lot to both of you! I really appreciate it.

Quote: randall jones wrote on Mon, 30 January 2006 08:09

> OK, Wesley Smith has compiled a 2up.jit.centroids external for 

Thanks a lot to both of you! I really appreciate it. 

Quote: randall jones wrote on Mon, 30 January 2006 08:09
----------------------------------------------------
>
> OK, Wesley Smith has compiled a 2up.jit.centroids external for  
> Windows.  Thanks Wes!  Big up community!
>
> -Randy
>


Ok. That might a bit late for this project, but let us all know when you are ready to release it.

Quote: bakercp wrote on Sun, 29 January 2006 21:31

> You know, I'm not sure when it will be working. My goals is to have 

> I have positive preliminary results with test data (i.e. one blob per 

> object with no merging or splitting of blobs ... a sort of closed 

> world scenario), but it becomes more complicated when one is required 

> to intelligently group broken blobs with holes, etc.

Quote: bakercp wrote on Sun, 29 January 2006 21:31
----------------------------------------------------
> You know, I'm not sure when it will be working.  My goals is to have  
> it working within the next month.
>
> I have positive preliminary results with test data (i.e. one blob per  
> object with no merging or splitting of blobs ... a sort of closed  
> world scenario), but it becomes more complicated when one is required  
> to  intelligently group broken blobs with holes, etc.
>
> I'll keep you updated.
>
> Christopher


cv.jit.label question