View Full Version : Motion Capture

02 February 2005, 06:15 AM
Hi All,

I suddenly (and I'm not sure how it happened) got interested in doing my own motion capture. I figured I could put little LEDs all over my body, film myself in a dark room, run the video through a software application, and spit out a bunch of 3D vertices. Then these vertices can be imported into an animation application and linked to a simple 3D character.

I know, I know... I'm sure it's easier said than done. But, in a single night, I whipped together a little motion tracking program. I turned on my webcam, turned the lights off in the room, and used the backlight of my PDA to get the tracking program's attention. I was able to get my program to draw (in real-time) a red box around the bright little PDA screen as I moved it around before the webcam. I'd say that's a start.

Translating video of "little white dots" moving around would actually require two video feeds shot from different angles. Then the "little white dots" could be paired up from each video feed and a vertice in 3D space could be calculated.

Okay, I'm afraid I'm going to have a lot of people ripping into me on this one, but... what do you think? Plausible?

02 February 2005, 05:03 PM
Wow man. That's an awesome idea :) I don't figure a reason why it couldn't be possible though you are probably going to run some problems with combining the two different angles and there can also be a problem with the illumination and reflections form the different leds. Maybe you just need to use very low voltage to get them mat. I wish you good luck with this :)

02 February 2005, 07:26 PM
that sounds similar to what's being done with optical mocap systems these days. normally you'd need more than two cameras though, also, the main problem seems to be how to handle the data when certain sensors are being obscured by body parts.
anyway, nice try.

however i imagine it could be somewhat difficult to act out a performance in a pretty dark room. ouch! ;)

02 February 2005, 08:00 PM
Hi All,

Thanks for all your interest! Yeah, I knew the part about the "little lights" being obscured by body parts would be a problem. But I'm thinking that when that happens, it only happens for a short time and at least one of the cameras will be able to see it. Then with some fancy algorithms, maybe the position of the lost vertice could be predicted. We'll see.
Thanks again for the input!

02 February 2005, 05:40 PM
the way you were doing mocap is, like already pointed out, the same way optical mocap is actually done these days.
the differences are that, to allow the actors to perform in decent light conditions, light absorbing suits and refractive markers are used; you need at least 4 cameras for most movments (it's convenient to always have a marker on at least 3 cameras, 2 is the bare minimum but often not enough) and last but not least...

how will your application know what markers on a camera correspond to what other markers on the other cameras' footage? ;)
unless you are ok building the LUTs by hand once you get all the footage in (cheap but very boring and long winded) you will have to start working with markers that can be read at different frequencies etc. and that brings you basically into a full fledged mocap system like we already have.

realtime 3D mocap with normal cameras alone, and setup in a casual environment, will be very limited, if possible at all.

many smart people worked on various mocap techniques, if the simpliest ideas had to be overcomplicated a bit to really work there's a good reason. But by all means go on with your experiments and have fun, you'll learn tons, but don't think you were the only person who had such brillant yet simple ideas ;)

02 February 2005, 12:39 AM
Once I had the bright idea that it ought to be possible to build a home brew 3d laser scanner. Didn't get beyond attempting to search for some sort of laser though. ;)

02 February 2005, 01:12 AM
your idea is fine, and do-able, you would be amazed at the similarity this type of project has with some of the robotics research being presented at conferences all over. This may be an area you would want to investigate. Most universities subscribe to all major scientific conference proceedings. usually these are offered in electronic versions these days.

You may find that a lot of this work is way beyond what your trying to do, but it never hurts, and you may even find a good book or two that gives algorithms that would help you refine locating the light source in 3D.

CGTalk Moderation
02 February 2006, 02:00 AM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.