Kinect technology thread

That's very promising. I presume this is using the full-res 3D feed, rather than the quarter-res feed XB360 has access to? They're also standing half the distance away, so the actual image resolution is going to be something like 16x what Kinect gets with a player at playing distance. It'd be worth MS pursuing this sort of interface though.

 

Sure, and that's even what Ubisoft said. evidently though, what they are doing is simpler than Kinect's full skeleton tracking as it's lower lag, and part of that is because they don't have to do full body tracking because they aren't puppeting an avatar. By going with with Depth Blob avatar, they are making savings. That's all part of using the available data streams to best effect to reach whatever targets your game has.

 

Been following this discussion (since it's taking up so much of the thread), but now that bkilian has added a real world test, it still doesn't sway your opinion?

 

I wasn't talking about the tracking I was talking about how it's displayed. meaning that because it's your image on screen doing whatever you do the system don't have to use the tracking points to recreate your movements on screen with a avatar.

 

Ah, ok, got you now. Yes, all it's displaying is the depth image of you as well as the color image overlaid (face for example) from the color camera. In some activities it does show basic skeletal points also but only in the slower yoga/tai chi/stretching activities.

Regards,
SB

 

Another kinect minority report style video.

http://vimeo.com/17180651

 

What opinion is it supposed to sway? That not having full skeleton tracking is no quicker or easier to process than having full skeleton tracking?

 

Maybe I should have said the theroy or assumption which was the topic of discussion instead of opinion. If not then disregard. I found it helpful in drawing out an understanding.

 

easy to give publishers more incentive to develop games due to lowered games saturation. ms wants big publishers so that they can help out with marketing, indie developers wont.

as for the usb limitation for kinect i bring up another unlikely df article
http://www.eurogamer.net/articles/xbox-360-storage-update-the-flash-factor-article?page=2

seems that the maximum read speed off these usb devices is around 15MB/s (either because the interface or the OS limits it) anyways i use that as a base line for the following calculations

15 * 1024 * 1024 = 15728640 bytes per second

640*480*30hz = 9216000 pixels per second

15728640 / 9216000 = that only leaves 1.7 bytes per pixel.

===

320X240*30hz = 2304000 bytes per second

15728640 / 9216000 = that only leaves 6.8 bytes per pixel. (enough for 32 bit precision per pixel with 2.8 bytes leftover minus overhead)

this is probably pretty baseless but maybe it explains why the depth sensor resolution is halved. due to the limitation i dont think that kinect actually sends both the rgb and depth to the xbox i think it only does one or the other. or swaps between the two every frame.

 

It does both though, as we have the indie's compositing the live feed with the 3d data to prove it. Plus you need to add audio into your bandwidth consumption. Don't know if that's all four channels of the isolated sound. Should be the latter to save BW and explain the presence of some of that processing power in there, which would necessitate instructions from the host platform as to what spacial position to track. Unless that's done by face tracking internally.

 

New vid using 2 Kinects.
This guy seems to be one of the most active Kinect homebrew developers. He admits that it is still pretty rough, but figures it can improve with some calibration.

http://www.youtube.com/watch?v=5-w7UXCAUJE&feature=player_embedded

I'm curious as on the bottom left of the screen it displays fps. It shows an average of ~70+fps. Is that capture coming from Kinect or just the screen display refresh?

 

They just need to extend collision detection beyond the block. So that if you use the wrong part of your body you collide with an invisible area.

 

That's his redrawing method. Looks very impressive, although highlights how blacks can throw the IR out. Next-gen tech will be pretty incredible if they can get past the noise issue.

 

but thats through pc where the usb is running closer to full speed. im not saying the kinect device cant transmit both simultaneously, im saying that the xbox usb port is not full spec therefore is subject to some constraints.

 

Your Shape superimposes part of the color camera feed on the depth image of you in the game.

It's difficult to tell if it's doing it in realtime or just using a static image as I haven't looked at it too closely.

It's a bit weird though as it doesn't do any perspective correction. So as the image (face for example) wraps around the 3D depth generated avatar, the edges stretch.

Regards,
SB

 

If they're doing all that, nothing. But if they only need track the hand and feet positions to determine contact points with the right limbs, they won't be using full skeleton tracking.

Reasonable point. If the game refresh rate is 30 fps, then you don't need any tweening. If Kinect games are 60 fps, they'd need to buffer one frame to tween. Do we have framerates for these titles?

 

Here is a fairly good video of Kinect tracking



Almost all the videos I find, the person is just standing in one spot. In table tennis, you could move side to side around the table, and forwards or backwards. You can even duck underneath the table with just your hands sticking above. All the videos the guy just stands there. It even tracks your legs while you're playing, so you can serve with one leg up in the air etc. It's too bad no one tries to show how it really works in these vids.