I did the tutorials for about 20 minutes and after that started checking out the examples included in the girlpower folder. Seeing how other even unrelated patches work can give insight on how you might incorporate some functions in you’re own patch.
For now I’m going to take a little break from multi-touch I’ve been working on this nonstop since december. I handed in my report after a weekend without sleep. Done with it!
I posted a question about calibration in the vvvv forum. This reply is quite useful:
joreg: [25.01.07 01:19 UTC ]
halo joobs.
there is no builtin image undistortion (may i say yet). for now we’ve successfully used a freeframe plug once that mimics a fish-eye effect. can’t remember its exact name. but it is one of the older freely available..i think it was one of petesplugins to be found here.
alternatively you can do arbitrary undistortion by putting the video on a grid. afterwards use a DX9Texture->AsVideo combo to get the renderers result back into a videostream for further tracking..
this could turn out to be slow, but you could try to work with rather slow resolutions for tracking (by setting a low resolution on the DX9Texture node).
also try setting the AsVideo nodes “Clock” pin (seen via inspektor) to ‘None’. this could also speed things up.
a little late, but heres my opinion about the pros and cons of using vvvv only or vvvv and processing: there aren’t any. it just depends on what you are planning to do with it. if it is “just” some visual output depending on the tracking data, vvvv is fine. but once you start building something more complicated, like an interactive table with a multitouch interface which for example gets images and other content from a database, maybe even over the network and has tools for interaction, you’re better off using java (processing) or C++, since you can more easily create custom functions. if you have no experience in using an OO programming language, vvvv is the best tool ever for rapid prototyping…
What kind of a send rate have you guys been getting with OSC- I was sending data from vvvv to macromedia director, and the framerate wasn’t so good and i’m not sure whether it was having a rubbish camera that was the problem, or whether it was Director not recieving the OSC fast enough, or a combination of the two.
Hey Goobs, how many FPS are you getting out of VVVV? I am getting no real lag from a laptop to laptop over 100mb ethernet. I would have to say your camera, because it is not that much data passing from VVVV to Director (not sure how well Director deals with OSC) Out of VVVV I typically get arround 25-30 frames out of my “VideoOut” node.
Also take note that VVVV takes almost all of my CPU (2.0ghz) when doing blob tracking. So if you are doing both on same pc, that could be an issue.
I think its more the camera. I’m not sure what the framerate was, but if you check a video of it working (http://www.andybest.net/vids/2007inst2.html), you can see the effects of the bad framerate.
I was using a dual Xeon workstation to do the tracking and drawing, so i don’t think that was the issue- vvvv was using <10% cpu power if i remember, but i was only using a cheapo webcam. I think i’ll try and get my hands on a zoomable security camera with low lux sensitivity, which should hopefully provide better results.
I think i’m gonna try processing too, although I don’t know java right now, but I know C, so i’m sure i’ll cope.
I guess the camera is the bottleneck here, on two computers (2,66Ghz and 3,2Ghz) I found out that the performance was good but the camera framerate sucked and pulled the rest down.
we came up with the idea to let the gpu do all the image processing, so you don’t need a second pc.
a gpu is build for these kinds of processes and the cpu is not
we looked into building shaders (coding to let the gpu do the math), which is at this point still a level too high for us.
maybe somebody else has more experience with programming shaders, this would be a major speed issue solved
there’s hardly any information about it on google about shaders in combination with this ftir project because probably not alot of people had thought about this solution
you got to have quite some knowledge about opengl and stuff....
this might be interesting to look into, because quite a few people encounter the problem of slow results because an cpu overload
Hey hey, i like the idea of using just shaders to do the blobdetection… since i am not a hardcore coder, i can only hope that somebody reading this is. maybe this issue should be dealt with in a new thread?
EDIT: sorry we haven’t seen the extra /clear message in the vvvv scetch, now it’s working.
Someone know why vvvv doesn’t kill an already released blob?
The problem is that if you touch than release and than touch somewhere else, vvvv is sending the same ID for the touches as if they were never released. vvvv is supposed to kill the ID inbetween the frames no touch is detected.
We are currently using Jens vvvv scetch.
okay, after looking through the code of the blob receiver in processing again, i could find out why the clearing didnt work… now i updated the sketch… i changed the order of the strings when plugging the message to the clearBlobs action… i wonder why it worked in the beginning…
Hi! First time trying use VVVV… Its hard for me, I not have exp in java or other programing. But I read F1 and even so can adding right node for getting video stream from cam and adding blob detection. But I still cant use background remover. Who know how use this node?
on attach:
blue - my worked correction
red - what do with this?
