Kinect for Windows SDK v1.6 Enables Seeing in the Dark | Hacker News Reader

Kinect for Windows SDK v1.6 Enables Seeing in the Dark

microsoft.com

58 pointsbenlower27 comments

I work on the Kinect for Windows team and would love to hear from anyone who is working on a K4W project or has an idea. My job is to empower developers to build amazing experiences so please hit me up if you have questions or need any help. @benlower or kinectninja at microsoft dot com.

Almost 2 years ago I worked on a project to create a real-time avatar of Air New Zealand's Airpoints Fairy. Used a Kinect to do live tracking of an actress who interacted directly with audiences (via video and audio hookups). She co-hosted an awards show and a big rugby tournament.

The Kinect worked amazingly well but head tracking was a little iffy. So we strapped a smart-phone to the actress's head and used it's compass/gyroscopes to provide fine head tracking :)

There is isn't a lot of footage on the web but check out: http://theflyingsocialnetwork.com/archives/1545

sounds pretty cool to have a real-time avatar. your description of the smart-phone strapped to her head reminded me of this classic hands-free device: http://ts1.mm.bing.net/th?id=I.4694782262707040&pid=1.7

You laugh but actually the phone was pretty much duct-taped to a base-ball cap ;)

I'm moving soon and will be setting up a new home office and one of the things I was thinking about is using Kinect as a sort of posture sensor at my desk, so if it detected me leaning forward, etc., it'd be able to throw a warning on the screen. Is there anything along these lines I should be looking at? I certainly don't mind writing it if I need to but I cannot be the first person with this idea.

i've not heard of that solution being done yet but it's a good idea. LMK if you work on it as i could use it to stop leaning back so much in my chair. question is do you do carrot or stick model? carrot would be something like for 30 min in a row of good posture i get a reward. stick would be that super-annoying sounds are played (alarm, etc.) until i correct the posture.

I would probably do the stick approach via something like a translucent overlay. I'm not sure the carrot approach makes sense to me personally because the rewards (no neck pain, etc.) are pretty intrinsic already.

I will definitely throw it up on github if I don't end up finding something else that already does it.

On a similar note, I have also been looking for something I could use to move my cursor between my monitors. I have a sort of a half-hexagon arrangement using 3 displays and it would be nice to be able to center the cursor on whatever one I am looking at.

I think there are applications out there that sort of do this, though, so I am hoping I luck out.

"My job is to 'empower developers to build amazing experiences' so please hit me up [...]"

That looks so much better in quotes, because in my experience only HR, marketing and pointy haired bosses speak like that.

haha. nice. i guess i was just outed. thought the haircut would keep my identity secret a bit longer.

I'm working on a research project where we use Kinect (+ accelerometers) to detect one person's activities.

We're using the skeleton provided by the SDK and it doesn't work very well when the person is not facing the camera (which is understandable since it was originally designed for games). Also, the SDK skeleton doesn't seem to have any idea about morphological constraints (e.g. impossible arm rotations). Do you plan/is it possible to improve the skeleton tracking part ?

i'd love to dig into this more to see if there's anything we can do today to help you with this scenario and, if not, what we might do in future. will you please email me at kinectninja <at> microsoft?

Nice. Purchased a Kinect a couple of days ago and failed to get it up and running with SDK 1.5 in a virtual machine. Thanks for making it work in virtualized environments.

Any chance to get my hands on KinEtre?

Finally. OpenKinect has had these features for quite a long time. =)

I'm happy that Microsoft is officially supporting these features - and I'm looking forward to seeing what gets built.

Thanks :-) Developer & customer feedback is a big deal for us and we've been working to add new capabilities that people want. Anything else you can think of that you'd like to see us do or enable in the SDK?

I would like the source!

not sure if you're being serious or not :-) while you can't clone our source code's repo, there are some ways that you can inspect things at a lower level: 1) debug .NET framework source (http://msdn.microsoft.com/en-us/library/fc12e472-ac6a-4e77-8...) 2) use a tool like Reflector to inspect our managed layer that we expose in Microsoft.Kinect.dll. We also publish the source code for a number of great samples (look in our Dev Toolkit)

So you don't provide source code, or condone linux access. Why is that, what do you fear?

That's not a hostile question at all! /s

A little guy questioning a humongous pile of money empire (I doubt they'd feel it).

Yup! I am being serious. Can't is just a word, and I don't think anyone would have any real long standing moral qualms if the source were to somehow suddenly appear in someone's mailbox in France! :D

Higher resolution image capture for applications dealing with face recognition would be awesome. Is the team holding off on face recognition until a higher resolution camera/kinect is in place?

The camera resolution is fixed by the hardware. I think they'd have to wait until a better Kinect is built.

That said, you COULD go the way of several hacks I've seen - marry a high resolution external camera to the Kinect rig, do the correspondence, and use the external camera as your face recognition input.

The Kinect (at least the version for the Xbox 360) is capable of providing 1280x1024 Bayer-encoded RGB images at a reduced frame rate.

james: would love to better understand your scenario...might be more guidance we could provide. pls email me more details if you want (kinectninja at microsoft <.> com).

that said, we do enable you to use the infrared feed and marry that to a hi-res external color camera.

Why don't you provide object recognition on a large range of classes, like http://www.image-net.org/ for example, and millions of specific objects + ocr, like what google goggles provides http://www.google.com/mobile/goggles/

Your sdk is slow moving and limited for all the resources that your empire has at it's disposal.

No need to have a hostile tone here. We have a guy from MS here. I'm sure he doesn't make policy decisions, like what ends up on the SEC report.

Do note, that the SEC report the last I checked, included the biggest competitors as Linux and the GPL. The last they'd give is source for the free masses to use.

Yeah, but they are getting free marketing on here, it's not exactly a bootstrapped startup.

Reply on news.ycombinator.com