Source: MS quadrupling Kinect accuracy
Finger movement and hand rotation possible.
Microsoft is working to improve the accuracy of Xbox 360 motion-sensing add-on Kinect so that it could detect finger movement and hand rotation, Eurogamer understands.
Microsoft's Kinect team is said to be working "very hard" on a switching or compression technology that will allow a greater amount of data to pass through Kinect to the Xbox 360 console.
Kinect features are dictated by firmware so that they can be added and upgraded over time.
The depth sensor used by Kinect is also dictated by firmware – it is currently set at a 30 frames per second limit and a 320x240 resolution limit.
At a 640x480 resolution, however, Kinect could begin to detect fingers and hand rotation – an effective quadrupling of its accuracy.
The issue relates to the USB controller interface, Eurogamer was told. It is capable of around 35MB/s, but it only uses around 15/16MB/s.
This artificial limit is in place because multiple USB devices can be used at once on an Xbox 360. But Microsoft is working on a technology to allow greater throughput in this regard, Eurogamer understands.
If Microsoft achieves its goal it could double the spec of Kinect's depth camera with a simple dashboard update.
Microsoft had not responded to Eurogamer's request for comment before publication, but Digital Foundry's Rich Leadbetter described the potential accuracy improvement as "eminently doable".
Microsoft would need to "disable or lower throughput of game installs running from USB flash drives to free up additional bandwidth," Leadbetter said.
"All eminently doable though bearing in mind that Kinect 'only' needs 20MB/s for full res from both cameras.
"The resolution coming out of the depth camera via PC is indeed 640x480, but it is uncertain just how accurate the camera's sensor is.
"Additionally, processing four times as many depth pixels could slow things down more."
Last month Anton Mikhailov, a software engineer at Sony Computer Entertainment America's research and development department, told Eurogamer Sony turned down Kinect's 3D camera because of the limitations of the tech.
"In reality, the 3D cameras we surveyed and what Kinect ended up using, they're 320x240 resolution, so when you're talking about tracking fingers, or even tracking things like the rotations of your hand, you're working with 10x10 pixels," Mikhailov said.
"It's very hard to get anything useful out of it."
Mikhailov doubted Kinect's capacity to create a decent Star Wars game "because there are so many ambiguities, and it's nearly impossible to track the angles of your wrists".
If Microsoft achieves its goal of improving Kinect's accuracy, however, Mikhailov could be proved wrong.
You may also like...
-
Asura's Wrath Review
-
In Theory: How iPad 3 Breaks the 1080p Barrier
-
The Rise and Fall of Sega Enterprises
-
GAME to close 35 stores
-
AMD in "hush-hush effort" to put graphics chip in PS4
-
The Essential PlayStation Vita
-
Gameplay signs off with £5 Gamestation.co.uk voucher
-
Syndicate Review
-
BioWare: we finished Mass Effect 3 before working on DLC
-
Ridge Racer Vita Review
-
Sony: 3G gives Vita "immediacy"
-
Battlefield: Aftershock pulled from App Store
-
Assassin's Creed: Revelations The Lost Archive DLC details
-
Amnesia: A Machine for Pigs announced for PC
-
Dating site for gamers launches in the UK
-
GAME: "we can't stock absolutely everything"
-
Can SSD Upgrades Boost PS3 Performance?
-
PlayStation Vita: where's the cheapest price?
-
Guild Wars 2 open beta sign-up begins
-
BioWare unboxes Mass Effect 3 video
-
FIFA 12 gets January transfer window squads update
-
BioWare blasting Mass Effect 3 copies into space
-
Japan chart: Strong debuts for Binary Domain, Theatrhythm
-
PS Vita: Sony defends Uncharted, FIFA price, explains expensive third-party digital games, reveals larger memory cards are coming
-
PlayStation Vita games: where's cheapest?









Comments (58) Latest comment 4 months ago
Comments threads automatically close after 30 days, but please feel free to continue chatting on the forum!
Comment below viewing threshold Show
Comment below viewing threshold Show
Either way it is rather embarrasing for the Sony engineer to be saying things that DF could prove incorrect. Unless during the process of making and developing Kinect Microsoft worked on a new and better camera than the one Sonyman saw.
Comment below viewing threshold Show
Comment below viewing threshold Show
Depends entirely how close you are to the camera.
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
We'll probably eventually see some decent games for Kinect that don't need the extra resolution anyway.
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
There is a communication bandwidth issue as one problem, but there is also a memory bandwidth issue if this additional data will be processed in main memory by the Xenon cores rather than on the GPU.
But if they do decided to use more GPU resources then the games will suffer in other ways, so I'm inclined to believe they won't be able to do it well with games; even if they could achieve it in the future.
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
Kinect haters must be feeling pretty embarrassed right now,first sales,now this
lulz
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
@ captain_Carl - So you'll continue to read articles about Kinect, and post how you don't give a fuck, all whilst not caring at all about it? Riiiggghhht.
Comment below viewing threshold Show
Comment below viewing threshold Show
You feel sorry that parents won't buy 18-certificate Call of Duty for their kids for Xmas?
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
I bet good money that is *exactly* what they want.
Comment below viewing threshold Show
Comment below viewing threshold Show
Sounds correct; and will most likely double or quadrupling the processing requirements, just like it does if you texture map with a texture that is double in both dimensions; but people don't seem to like the harsh reality, that a regression analysis with 4 times the samples, requires four times the computational work.
Comment below viewing threshold Show
Comment below viewing threshold Show
Just because one company could not figure out a solution to a problem doesn't mean another will not. A lot of companies have rejected ideals because they could not see the solution only to have another company come to market (Wii). Sony and MS definitely understand this concept pretty well.
On another note, isn't the original 360 holding back Kinect. The 360 s has the power supply and USB cable in one. I am sure that USB port is dedicated to Kinect and thus do not have a bandwidth issue.
Comment below viewing threshold Show
Your lulz are hollow and pathetic.
Comment below viewing threshold Show
Not really.
Right now there is enough material to laugh for months,but I'm not a mean person.
lulz
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
You know something. If MS can really tap this market, I am sure you can lolz all you want. This is a huge and very untap market. This could also be an entry point in Japan that MS is always looking for (too bad about the space issue). Teen/ girl gaming market is huge in Japan and even the US. It's very much untap and the company that could leverage their device to hit that market will run away with sells out the ying yang (opps looks like Nintendo already have).
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show
I thought you liked Kinect?
Comment below viewing threshold Show
but some developers already said that internally, the IR sensor of the Kinect is higher than the projected IR image.
if you check out the Anandtech review, the guy stated that too:
"The projected image doesn’t change in time - it’s fixed this way. The IR CMOS sensor images this pattern projected onto the room and scene, and given the camera’s displacement a few inches from the projector, from the displacements in the semi-random projected pattern is able to back out the corresponding depth image. That computation is done onboard the Kinect itself, and it’s entirely possible (read: likely) that the IR sensor inside the Kinect is higher than the 640x480 resolution of the resulting image."
Comment below viewing threshold Show
Way to miss the point, dumbass.
The point is that it's capable of accuracy already, the titles used are irrelevant as the capability can be transferred into better games or games more suited to core gamers. If MS are indeed improving it that only adds to that point.
Comment below viewing threshold Show
the PSEye and PS Eyetoy are just RGB cams. they can only work by detecting pixel changes. because it works by merely comparing pixels, the conditions needed for it to work are very stringent. if you actually played eye games back in the day, you would know this. the lighting has to be spectacular, the clothing has to be a clear contrast, etc.
also, with only an rgb cam, it's nearly impossible to detect depth. that's why eye games only had motion in a 2d plane. you reach up to pop a bubble. you jump up and down. to detect depth in a sea of pixels is a logistical nightmare.
that is why the move uses a colored ball. that ball is a clear contrast to the environment. very easy for an rgb cam to pick up. they see a pink pixel coming off the ball and chances are the surrounding is not pink. that ball is also a fixed sized so the system can detect if the move is moving toward the screen or away from the screen based on the measured size. if the ball is smaller than expected, it's moving away. if it's larger, it's moving closer. that's the depth work. then you add in the sensors inside the move to detect orientation of the move controller.
to do what kinect is doing needs more than what an rgb cam can do. in controlled environments (which you never get outside of a lab), you can do minority report stuff with only an rgb cam. but the thing is you can't expect that in a typical home environment. people come in all different shapes and colors. no two people motion exactly the same way. there are so many different types of clothing. worrying about shutters or blinds in the background, etc.
kinect does none of this because it's motion tracking does not depend at all on an rgb camera. you can play dance central in complete darkness. if Dance Central was working the way Harmonix games worked on the Eyetoy, it would not work at all. Go back and play any eyetoy games in the dark. They don't work.
Comment below viewing threshold Show
Comment below viewing threshold Show
btw why in the artcle many "Eurogamer understands" ?
english is not my native language, i kind of confused with that....
Comment below viewing threshold Show
Comment below viewing threshold Show
english is not my native language, i kind of confused with that....
Non-cynically, Eurogamer have used these words because Microsoft are still hard at work trying to make this happen (believing they will) from their own R&D & adapting any open kinect techniques (running on PC hardware); so getting a firm quote to say they will achieve it on 360 hardware and the USB bandwidth for Kinect will be changed hasn't been given corporate approval yet, but it will shortly.
I'm actually really interested in the technical problem of how I might interpret this data for motion recognition in real-time, and would gladly take a free kinect to play with on PC for development fun. But as yet, I still can't quite generate one of those, “wouldn't it be great if you could ...” ideas for games with a 3D camera; probably because I think accelerometer wands will always be faster to communicate a user's broader motions and can provide feedback with rumble and sound.
Comment below viewing threshold Show
Comment below viewing threshold Show
Keep in mind both Sony and Nintendo have turned down Kinect. Also, there is video footage of of Sony working with the Zcam (which was it's name back then). Also, what makes you think that Dance Central isn't using the technique brought up by Mikhailov? Dance Central has ZERO need for actual motion tracking, so why not cut down the overhead and latency by just matching key frames of the skeletal figure with the ones in the game? It's so much easier and is more logical. That's why even Dance Central is (in some ways) more accurate than games like Kinect Sports.
Also, what does bringing up Move have to do with anything? They (Kinect and Move) are two different technologies. If you're trying to bring this (the fact of it being only an RGB camera) up as a point to why Mikhailov doesn't know what he's talking about, you're mistaken, both Richard Marks (the creator of the Eye Toy) and Anton Mikhailov have both worked in both RGB and RGB + IR tech. But that's besides the point.
Comment below viewing threshold Show
I think Sony and Nintendo's engineers are too professional to get into smack talk, were their professional reputations are concerned; Nintendo tend to focus on making choices that are “correct” to compliment their family and child friendly outlook, and Sony tend to focus on making “correct” choices from a science and lifestyle perspective; ie “correct” for educated people.
They've both said that 3D cameras are good, but professionally substantiated reasons why they weren't correct for them to use in a variety of gaming experiences.
Microsoft on the other have been known to use the expression Embrace, extend and extinguish and have mostly been successful from being closely tied to an architecture that people believed could result in Electronic Armageddon in 2000 with the Millennium Bug. So, taking the correct choices from a science perspective has probably been less important to them than market share.
Comment below viewing threshold Show
Save your cash... Kinect's problems are not just the lag and poor recognition problems, it's far bigger problems are that it's VERY limited in the type of game it can work with. How many shovelware jump and wave your arms about can you put up with....
Comment below viewing threshold Show
Comment below viewing threshold Show
Maybe, but the bandwidth calculations above, “needing 20MB/s for full res” are surely wrong, if we are meaning 24bit RBG camera, 16bit IR depth (40bits total) @ 60fps and 640x480.
As that actually needs:
40*60*640*480/(8*1024*1024) = 87.89 MB/sec bandwidth, which probably exceeds the 35MB/sec they said the 360 USB controller provides even if they can get a real-time lossless compression working; as that is a big compression ratio needed.
Comment below viewing threshold Show
You can probably knock that number in half, I don't remember 60fps being part of the equation. Equally, I'm not sure that all games are going to need both the RGB and the IR feeds.
Comment below viewing threshold Show
Who am I kidding Kinectimals is awesome!!!
Haters keep hating.
Comment below viewing threshold Show
A small point but your 87.89 mb/sec calculation assumes no compression what so ever and, if they use 16 bit colour at 30fps, then the calculation comes out at 35.16 mb/s, again assuming no compression. Now use the depth data as a mask for the colour data and lossless compression becomes unnecessary.
Comment below viewing threshold Show
Comment below viewing threshold Show
A small point but your 87.89 mb/sec calculation assumes no compression what so ever and, if they use 16 bit colour at 30fps, then the calculation comes out at 35.16 mb/s, again assuming no compression. Now use the depth data as a mask for the colour data and lossless compression becomes unnecessary.
Having study signal processing years ago, I don't think 30fps is enough samples per second to accurately capture finger movements(with 33ms between samples); so just quadrupling the quantization levels doesn't fix the second problem of sample-rate.
Adding resolution would certainly help them, if they try to extrapolate more frames with subframe interpolation, but then the processing demands to do this and delay from detecting the user motion to on-screen action will most likely get even bigger, using general purpose CPU core architectures(that max-out at 60% processing efficiency due to cache misses)
There is a big difference between people saying why they believe a product doesn't work fully as intended (technologically), and just hating on a product, but it seems like people who get fulfilment from Kinect games and are wishing for it succeed don't seem to differentiate the two types of comments.
Comment below viewing threshold Show
Comment below viewing threshold Show
Comment below viewing threshold Show