Realistic Human Motion from Monocular Video Sequences

That's pretty awesome, now Terminators will be able to blend in much easier by mimicking local human movements.
 
Impressive. Hard to tell how accurate the leg separation was captured in a couple of those examples, though.

It certainly seems good enough for some videogame purposes. I wonder if the post-capture adjustments necessary for realism would make this better than traditional mocap for feature films.
 
I don't want to sound like a Jerk and I hope I don't. I though this was exactly what Nata... Kinect was doing, knowing the general human structure, watching the video building an outline then putting the skeleton in there?
 
I don't want to sound like a Jerk and I hope I don't. I though this was exactly what Nata... Kinect was doing, knowing the general human structure, watching the video building an outline then putting the skeleton in there?

You don't sound like a jerk, this is pretty much what Kinect is doing.

It is a very cool and cutting edge technology. Kinect is easily the most cutting edge entertainment device you will be able to put in your living room.

I'd love to buy one... if they weren't charging $60 for XBLA quality games.
 
Thanks IRSmurf, I just hate it when people start saying things that let people know that they obviously don't get it. I still think its great that they figured it out.

I remember seeing a program (I don't think I downloaded it) a long time ago that could analyze video for scene detection and record from a video cam. So you could connect almost any video cam to the computer and let it run and see who has been messing with what while you were gone and only get the video from when there was action.
 
Pretty cool stuff. I guess that means Snoop Dogg won't have to put on the goofy suit capture suit next time someone wants him in NBA2kJamLiveOMGBBQ. The thing that always irritated me about games that used that sort of body motion capture was the falling. Everybody here remembers the falls of Tony Hawk 1 through Infinity right? You're moving full speed, you screw up and you slide on your stomach down the half pipe whilst in full standup position. Yeah because whenever I fell off half pipes, I managed to retain my full rigor.

Side note. This program would be great for backyard magicians/entertainers that specialize in balloon animals.
 
Thanks IRSmurf, I just hate it when people start saying things that let people know that they obviously don't get it. I still think its great that they figured it out.

I remember seeing a program (I don't think I downloaded it) a long time ago that could analyze video for scene detection and record from a video cam. So you could connect almost any video cam to the computer and let it run and see who has been messing with what while you were gone and only get the video from when there was action.

There are tons of apps to do that. I use Supervisorcam to record video, and have used it to record an employee 'shopping' in our tech room after hours. I capture from a usb to composite video dongle.
 
There are tons of apps to do that. I use Supervisorcam to record video, and have used it to record an employee 'shopping' in our tech room after hours. I capture from a usb to composite video dongle.

HAHAHA my dad used to call it shopping when we'd take pens and pencils and all sorts of shit from his desk when i was a kid. I forgot about calling it that. Can't wait to use it on my kids!
 
very cool. looking forward to seeing implementations in the future. this is a lot like a tech shown at e3...i forget the name
 
Alright, I'll play the part of the troll here...cause I don't get what's so special about it and was not impressed at all. Obviously I missed something because this looks like something that would be cool maybe 20 years ago. I watched the whole thig assuming that at the end they'd show me something that blew me away...but it never came.
Someone please set me straight.
 
Alright, I'll play the part of the troll here...cause I don't get what's so special about it and was not impressed at all. Obviously I missed something because this looks like something that would be cool maybe 20 years ago. I watched the whole thig assuming that at the end they'd show me something that blew me away...but it never came.
Someone please set me straight.
Traditional motion capture involves placing actors in leotards either covered in reflectors and filmed with at least two cameras, or covered in sensors and recorded in a magnetic field. Obviously, you need the actors to go through the motions while wearing the getup, recording them in that specially-equipped studio.

This technology can use the feed from any old web- or handi-cam, if you want... or retroactively on recordings made 20+ years ago. Anyone care to mocap some Astaire and Rogers? :)
 
I don't want to sound like a Jerk and I hope I don't. I though this was exactly what Nata... Kinect was doing, knowing the general human structure, watching the video building an outline then putting the skeleton in there?

Kinect does this in a very rough and raw form. It's taking movements and using them to cause action in a game. It's just not really trying to motion track you like this. It more or less takes basic arm and leg motions and maps them to key presses like punching your right arm will equal a press of the a button which the a button is setup to punch the right arm. If you look at the yoga app/game that they demoed at e3 you will see the raw orange data read that the Kinect system is rendering. If you watch it you will notice that you don't really see much of the body in detail just the arms and leg. The yoga app I'm sure just watches you make lines on the screen with your arm and legs and interprets that as correct or incorrect movement. It just doesn't go as far as knowing where your knee is and if it's bent at a 30 degree angle like this is doing. Also this app only seems to make key framing easier. It doesn't do everything all by itself.
 
Looks like something that can be packaged with a 3D modeling software. Imagine if you can generate your motion data from a basic home camera.

Natal is guesstimating your movements, this is more precise and has tools to let you fine tune how your motion data is generated.
 
I have been waiting and watching similar "tech" being researched at universities very closely for years. Thats all it ever is. Research. Yes Kinect will bring similar tech to the home. But what I want, and have still been waiting for, is CHEAP full body mo-cap for 3D packages. There are solutions to do this similar tech, but the minimum buy-in is around 5 grand. And a "proper" mo-cap system will set you back 30 grand.

So for now, that leaves the "hobbyist" 3D artist (like me) to still have to hand animate which is not only time consuming but woefully unrealistic. I've had some success with facial video capture but nothing for full body stuff. Still waiting. :(
 
So for now, that leaves the "hobbyist" 3D artist (like me) to still have to hand animate which is not only time consuming but woefully unrealistic. I've had some success with facial video capture but nothing for full body stuff. Still waiting. :(

In the making of "Avatar" they showed a head rig with a camera pointed at the actors face. I can see a lot of home modelers copying that and hooking up USB webcams to a football helmet :) This one can handle the body motion.

It may not have the same freedom as running around a football sized room, but you often don't really need that.

Bending over a table to pick up something was a challenge. The only way i made that was by recording myself with my cellphone, and then moving the rigged model frame by frame to match it. That was a pain in the neck even before i added the finger movements. :rolleyes:

http://www.youtube.com/watch?v=4Pjr-knQEm0

Walking around, putting his gear together, all those hours i could have saved if i had this:(
 
Back
Top