PDA

View Full Version : Cheap Motion Capture



Nakoruru
11-27-2002, 06:19 PM
I know that this is not precisely on topic, but there are a large number of hobby and student graphics programmers on this board and going to a motion capture forum would skew the answer to my question. Anyway, if I decided to do this, it would be worthy of a link on the front page of opengl.org.

I have been researching the feasability of building an optical motion capture setup which could be built for less than $400 in hardware costs. The whole thing could fit in a duffle bag.

My question is, would anyone be interested in this? (a sub 400 dollar mocap setup) I am asking because I hardly want to do something that no one thinks is cool ^_^

uzu_manga
11-27-2002, 10:29 PM
400$ in hardware you say... hmmm... How advanced is the equipment? And what is included?

Dr Doom
11-28-2002, 05:55 AM
do you want to be able to fully process the data on site? in which case, I guess you're not including a laptop in that $400.

To do it optically... you'll need a lot of brightly coloured ping-pong balls (cheap) and a couple of decent cameras (not so cheap)

I knew a guy looking into using webcams for this purpose. He reckoned it was doable, but he couldn't be bothered doing the serious coding required (i think he uses a 3 miniDV camera setup now - and even then he does the processing in a fairly lengthy offline process)

I dont know about doing it with radio or sound type equipment... tuning in to each sensor's frequency and measuring it from microphones/whatever... sounds complicated to me, but there might be a tandyhead here who could do it easily http://www.opengl.org/discussion_boards/ubb/wink.gif

That's for 'dot' based capture. If you're only after basic human movement, making your own strap on joint sensors shouldn't be too hard. I hear that systems like that, based on tensile reactions to movement often lose fidelity quickly (as the material stretches)

I'd go with the optical solution myself. It's surprisingly easy to calculate 3space position from even a pair of images, given a known camera calibration matrix and some decent reference points in the scene... which gives me another idea... but one unrelated to this thread, so I'll shut up now...

I'd be interested to hear how you go

Dr Doom
11-28-2002, 06:00 AM
do you want to be able to fully process the data on site? in which case, I guess you're not including a laptop in that $400.

To do it optically... you'll need a lot of brightly coloured ping-pong balls (cheap) and a couple of decent cameras (not so cheap)

I knew a guy looking into using webcams for this purpose. He reckoned it was doable, but he couldn't be bothered doing the serious coding required (i think he uses a 3 miniDV camera setup now - and even then he does the processing in a fairly lengthy offline process)

I dont know about doing it with radio or sound type equipment... tuning in to each sensor's frequency and measuring it from microphones/whatever... sounds complicated to me, but there might be a tandyhead here who could do it easily http://www.opengl.org/discussion_boards/ubb/wink.gif

That's for 'dot' based capture. If you're only after basic human movement, making your own strap on joint sensors shouldn't be too hard. I hear that systems like that, based on tensile reactions to movement often lose fidelity quickly (as the material stretches)

I'd go with the optical solution myself. It's surprisingly easy to calculate 3space position from even a pair of images, given a known camera calibration matrix and some decent reference points in the scene... which gives me another idea... but one unrelated to this thread, so I'll shut up now...

I'd be interested to hear how you go

Nakoruru
11-28-2002, 09:39 PM
No fair asking technical questions while I am in the research phase ^_^ The goal is to have a working system with acceptable drawbacks for less than $400.

I have an idea of how I will do it, but I do not want to talk details until I have either succeeded or failed. I have a feeling that part of me succeeding will have to do with not being told "it can't be done"

I do believe that the software that I will have to write is the -number one- factor, meaning that I intend to write software that works with 'cheap' cameras (i.e., middle of the road webcams). But, if it works for these, then someone could opt for more expensive equipment.

Of course, I am assuming that someone doing this has a fast computer, or they would not be doing graphics work. I am even considering some ways to use opengl with "GeForce 3" level features to accelerate processing of video frames (vertex programs to cancel lense distortion, fragment programs to filter the frames, automatic mipmap generation to accelerate finding markers, etc).

If you did want a dedicated box, I imagined using one of those tiny shuttle computers with a 4 port firewire card, an ATI 9700, and a 180GB harddrive. But is just a daydream. I've got to write the software first.

I am going to start experimenting with my MiniDV camera and a whole lot of other less expensive cameras...

Anyway, that is all the technical details I am willing to share. My question is only "Is anyone interested in doing motion capture in their basement?," or, "Is anyone interesting in doing motion capture planning and prototyping at the office without having to reserve/rent studio time."

What would someone expect from such a system? I do not want to build someone no one wants ^_^

kansler
11-29-2002, 01:14 AM
For real cheap MoCap I suggest you'd take a look at Visual Marker. Works decent.
http://www.visualmarker.com/

Dr Doom
11-29-2002, 05:43 AM
I'm interested!!

I'm actually quite interested in doing it outside, not in my basement!

(My basement is my bedroom, so it might get a bit crowded...)

Of course, doing it outside would lead to all sorts of other filtering when clouds change the quality of the light etc... basically what I'd like to have ideally is a system where, with enough markers external to the object in question, you could get a) mocap data and b) camera motion data.

It would be great to be able to chuck a few sensors around the scene (preferably easily removeable ones) and then matchmove automatically... oh, the time it could save!! http://www.opengl.org/discussion_boards/ubb/smile.gif

Nakoruru
11-29-2002, 06:51 AM
Something about Russian software really scares me ^_^ (refering to VisualMarker)

I looked at that software and was left wondering why playback was so slow on a P4 1.8Ghz

I didn't investigate it much further, but my first impression was that I could make something better.

Re: mocap outside, With the changing lighting conditions outside, subtracting out the background would become a real problem. Actually, the fact that the room I have setup to test in has a window would be a problem. So would the software driven automatic white balance in webcams.

Ever wonder why something has not been done, then try to do it and find out why http://www.opengl.org/discussion_boards/ubb/wink.gif