Amit, I’ve put together a sample of the object loss due to camera distortion I was talking about.
This video shows use with a single object.

Initially I run it with both object and finger detection enabled.  This is why the ID changes as it does.  Then i turn off finger detection and just leave object detection on.

Anyway, let me know what you think.

grimjack2600 - 13 July 2010 10:15 PM

Amit, I’ve put together a sample of the object loss due to camera distortion I was talking about.
This video shows use with a single object.

Initially I run it with both object and finger detection enabled.  This is why the ID changes as it does.  Then i turn off finger detection and just leave object detection on.

Anyway, let me know what you think.

Thanks grimjack2600. Now I understand what you mean . I will take care of the distortion in next release. I will provide an option to undistort the image in the config.xml file and I will undistort the image according to calibration.  This never occurred to my mind though !

Also, in case you are using Finger tracking also, I would recommend to use a bigger object (say a mobile phone, so that the blob don’t come in the finger blob size range ) so that they don’t get confused.

great video thanks for sharing jack, and nice name too 2600

Great great work, tracked the objects and rotation perfectly on surface.
Although (at least for me) couldn’t receive UDP tracked data with TouchLib LCConnector :(

Hello,

I just compiled this object tracking version of CCV for testing on the Mac, so I thought I would share it with the community.

Its compiled for the 10.5 sdk, but I have only tested it on 10.6 using a PTGrey FireFly and the built in isight everything seems to work the same as the windows version, there was a bug in the code where the PS3 camera variable was only declared inside a #ifdef Windows but then called outside the #ifdef Windows, otherwise no major problems to compile.

I also compiled a second more “experimental” version which uses apples newer QTKit instead of the standard C Quicktime API. I only tested it with the PTGrey Firefly but the image quality is greatly improved over the standard version and supposedly it allows more USB cameras to work on OSX, however it does end up using about 20% more CPU cycles. Also I the camera settings dialog seems to be broken.

Its based on this addon I found on the openframeworks forum:
http://www.openframeworks.cc/forum/viewtopic.php?f=10&t=3555

-noah

slugmobile - 23 July 2010 12:03 AM

Hello,

I just compiled this object tracking version of CCV for testing on the Mac, so I thought I would share it with the community.

Its compiled for the 10.5 sdk, but I have only tested it on 10.6 using a PTGrey FireFly and the built in isight everything seems to work the same as the windows version, there was a bug in the code where the PS3 camera variable was only declared inside a #ifdef Windows but then called outside the #ifdef Windows, otherwise no major problems to compile.

I also compiled a second more “experimental” version which uses apples newer QTKit instead of the standard C Quicktime API. I only tested it with the PTGrey Firefly but the image quality is greatly improved over the standard version and supposedly it allows more USB cameras to work on OSX, however it does end up using about 20% more CPU cycles. Also I the camera settings dialog seems to be broken.

Its based on this addon I found on the openframeworks forum:
http://www.openframeworks.cc/forum/viewtopic.php?f=10&t=3555

-noah

Wow man , This is a great news ! Thanks for the work mate ! When next version is done, I will contact you. So that we can have the release for Mac too !

gl1tch - 22 July 2010 08:02 PM

Great great work, tracked the objects and rotation perfectly on surface.
Although (at least for me) couldn’t receive UDP tracked data with TouchLib LCConnector :(

Hi.
yes, people have reported problems with that. I am having a look into the code now.

Can you please verify if Only Finger Tracking gives such error or not? let me know.

Sure! No problem, looking forward to the next version!

Amit - 23 July 2010 02:21 AM

gl1tch - 22 July 2010 08:02 PM

Great great work, tracked the objects and rotation perfectly on surface.
Although (at least for me) couldn’t receive UDP tracked data with TouchLib LCConnector :(

Hi.
yes, people have reported problems with that. I am having a look into the code now.

Can you please verify if Only Finger Tracking gives such error or not? let me know.

Actually i was wrong, i got data to flash by UDP using LCConnector , but it only tracks fingers or objects at a time:

- if you only choose fingers it sends fingers id’s sessions
- if you only choose objects it sends proper object id session
- if you choose both objects and fingers it only sends objects id’s and keeps sending them even if object is not displayed

gl1tch - 25 July 2010 05:55 PM

Actually i was wrong, i got data to flash by UDP using LCConnector , but it only tracks fingers or objects at a time:

- if you only choose fingers it sends fingers id’s sessions
- if you only choose objects it sends proper object id session
- if you choose both objects and fingers it only sends objects id’s and keeps sending them even if object is not displayed

Thanks mate for pointing this out. This is a real bug which wll cause problems. Will be solving it soon I guess. Thanks =). Also the TUIO XML bug has been solved for people’s information =).

hey, great stuff you’re doing here. I am interested in the tracking process, does the tracker only differentiate between different sizes of rectangles or can it tell a triangle and ellipse apart?

Yes , right now I am checking with different sizes of contour binding rectangles. You can see this post where I elaborated why I chose this : http://amitsarangi.wordpress.com/2010/06/01/rotation-invariant-template-matching/. <-- This is much more faster and pretty accurate

But yeah , contour matching (as my original plan) is my next thing to research upon , my next semester academic project! May be I can get something efficient enough to integrate .

here a quick report from my first test
I spent a couple of hours tonight testing the 1.3.2 CCV COT on Windows 7.
It ran right out of the box, just need to change the “camera size” settings in the config.xml to hook my Firefly MV .
Then i tried to track a couple of transparent boxes that were laying around in the lab .
I did the test on an 47” LLP setup ( i built that blackbox last year with the Muchomedia team).
I had hard times @ the beginning cos’ i thought it was “contour detection “, in fact it’s more “ blob size “ detection .
Since LLP have great contours but no filled shape , i had to change completly my usual settings .
So i filled the empty shapes with a bit of blur & a lot of noise
Then It works nicely, the tracking is smooth ( even when the object spins around...)
Performance are perfect, it seems as fast as usual (60FPS).
Since i had to push the settings , the smallest object was a bit tricky cos’ it was almost the same size as the fingertips ...(as mentioned earlier by Amit))
I had no time to implement a small demo so no video ...
Congrats & thanks AMit for this great tool & new toy .

Everyone on this thread, I am very glad to announce that my work has been integrated with CCV 1.4 which has been released . Please check it out @ http://amitsarangi.wordpress.com/2010/09/02/ccv-1-4/. There are some major changes, so don’t forget to read fully before trying out . Any questions/suggestions , shoot !

This is great!  I’ve been testing fids and this works wonderfully.  Thanks again Amit (and all)!