Kinect vision

[ZackAlfakir]

Has anyone tried using the Kinect for vision processing? Is it more or less accurate than a camera with an LED ring? How did you manage to hook it up for use? Our current plan for it would be to have it send everything to 2 Raspberry Pi's hooked up together and then send it to the cRIO/DS. Any better suggestions?

[bob.wolff68]

My expectation would be that
a) The kinect is likely less accurate as it is intended for different purposes
b) It would be very cumbersome to integrate considering that you'll have to mount a laptop/netbook on the robot itself in order to use the kinect while the webcam is used directly by the cRio or remotely by a separate computer as the drivers station.

RoboBob

[defied]

I've been curious about this.

I have a kinect at home, and a Raspberry Pi, and have been delayed on testing how fast a 700mhz arm processor with 512M RAM can process info, and use it for determination.

I imagine if you had one pi to process and stream the video to the other pi doing the determinations, it could work.

This is the resource I have been looking in to.

http://openkinect.org/wiki/Main_Page

There would be some hand-compiling going on, I'm certain, so you'd want to be prepared for that. This may be an off-season project to see how feasible it is with the limited power of the Pi.


On to the Kinect:
The Kinect has a built in IR light (Which I think we all know, I'm just being reiterative here), and IR camera, and a color camera.
The camera is designed to work in depth and image recognition, so it wouldn't be less accurate, it would probably be a perfect use tool for robotics.
The IR camera picks up the IR light, and as the IR light fades, the kinect translates to depth of field/distance from camera. You could get a rough idea of how far you are away from an object (All though, I think a Range finder would still want to be included if you wanted to gauge distance on your bot). That and the color camera work together for determination and object recognition.

Just remember, it's more than likely going to tax the cycles on the Pi something fierce.

I hope this information helps!

Thanks,
D

[faust1706]

The kinect has been used by teams for the past few years. In 2012, Microsoft actually gave each team a kinect in the kit of parts. I'm not sure if it was used before that, but I wouldn't be surprised.

Here is a short tutorial on our website:http://ratchetrockers1706.org/vision-setup/

We used the O-Droid X2 last year, and are using the O-Droid XU with the asus xtion (it is much lighter).

The kinect is actually a really good piece of hardware. Many research groups have been using it because of how good is it and how cheap you can get them now, which I believe is in the ballpark of $50 used at a local gamestop.

I'm not sure how it will fair with the pi, but I'm sure it will work out fine.

Here is a nice paper describing the math of solving for camera pose estimation I referenced in the academic paper I wrote about the 2012 vision program: http://publications.asl.ethz.ch/files/kneip11novel.pdf