6DOF Positioning System - PS3 Eye and fiduciary markers

[xef6]

Hi Chriky, nice work!

I'm curious, are you more or less settled on the dynamic-camera/static-pattern setup? Is your goal to have 360° rotation for the user?

I tried to do mess around with this a while ago, also with a ps3 eye, but didn't progress very far. I was using a static camera and a moving pattern: http://www.youtube.com/watch?v=MCh4gQ7qszU. I never really thought of trying to get more than ~180° of rotation, because I kinda expect most games to have a way to turn (so sitting in a chair works). I'm curious what you had in mind.

[PatimPatam]

Hello again Chriky, really great job you're doing! In the end i decided to start my own tracking project (which i've been working on for the last couple months); i will have something worth showing very soon hopefully! I'm also using the PS3 eye, with C++ and OpenCV, but it's a very different concept than yours really..

Can I ask you something? Do you know (or someone else watching) the focal length of the ps3 eye lens in mm? For the 2 different settings if possible?

I know it can be determined using calibration functions with a chessboard pattern and all that, but i figured out i could save some time if someone already knows the values.. No luck with google!

Thanks a lot, and keep up the good work!

[MemeBox]

I don't know the values for definite but my automatic calibration technique generates an estimated focal length for the cameras (PS3 eyes) I am using. The calibration produces consistent focal length values and qualitatively correct calibrations, so I suspect that the values are correct. I will grab and post them here if you are interested...

[PatimPatam]

Thanx MemeBox, that would be great! (and sorry for stealing your thread Chriky)

[MemeBox]

I get about this value for the wide angle view:
573

But this is in pixels units, rather than millimeters. You can convert to millimeters if you know the ccd width, which I think can be found, do you need it in mm's?

For my application the precise value seems not to be a show stopper, values in the right ball park +-50 pixels seem to work fine. A good starting point for the focal length (in pixels) is 1.2 * max(width, height).

Let me know if you need the value in mm's and can't figure out how to get there.

I'm also sorry for the thread hijacking

-EDIT sorry just realised that this is without the IR filter which actually also does some focusing, so this figure may be different from yours. I would recommend downloading Visual SFM: http://homes.cs.washington.edu/~ccwu/vsfm/, using many images taken from different locations looking at a static scene with the PS3 Eye as input to the application. This will allow you to find the location that the camera was in when the pictures were taken and in the process the application will estimate the focal length for you (although still I believe in pixel units), you may have to dig through the directory structure within the application to find some log which contains the value, I can't remember, soz...

[Chriky]

I've got the values at home I'll post them tonight.

If people were wondering I had some trouble identifying the LEDs correctly on the full board with everything lit. I think the green LEDs are too dim compared to the white. Getting the colour of an LED wrong is a major problem...

Needs a little bit more experimenting unfortunately, but I haven't given up

[Chriky]

@ xef6

My current idea is something like this...


The PS3 Eye would be fixed to the backtop, and would point up at the ceiling at a slight angle. From my experiments this means that apart from crawling on your belly (but including natural crouching), the PS3 Eye can always see the ceiling.

You can use this to get 6DOF for your torso (because the laptop is fixed firmly to your back).

I would then use the 3DOF tracking from the Rift to get the orientation of the head and use a slightly more advanced version of Carmack's "head on a stick" method from D3 BFG to work out the position of the eyes relative to the torso, by using just the orientation. I think that there is actually a 1-to-1 relationship between the position of your head (relative to your torso) and the orientation of your head (relative to your torso), it is just not quite as simple as "head on a stick" because your neck bends.

For single user you could add a Hydra to the torso and get hands or feet (guessing the elbow and knee positions with IK) but unfortunately I want a multi-user solution which rules out Hydra because of interference between base stations.

You should be able to do something a bit like this - http://www.metamotion.com/gypsy/gypsy-m ... system.htm - like the UE3 demo on the General board that uses IMU on each joint and FK.

[PatimPatam]

@Chriky

Thanks, I would really appreciate it if you can. I'm especially interested in the low FOV value (red dot), but if you have both even better!

@MemeBox

Thanks again, yes the OpenCV functions require the focal length input in mm.. I will try your method if i can't find the values more easily!

[brantlew]

Nice. I'm glad you guys are working through this.

[Fredz]

Thanks, I would really appreciate it if you can. I'm especially interested in the low FOV value (red dot), but if you have both even better!

From http://en.wikipedia.org/wiki/PlayStation_Eye :
"The PlayStation Eye can be set to a 56° field of view (red dot) similar to that of the EyeToy, for close-up framing in chat applications, or a 75° field of view (blue dot) for long shot framing in interactive physical gaming applications.

[PatimPatam]

Hi Fredz, we were talking about focal length values on the previous posts, so I meant that I'm interested in the FOCAL LENGTH value for the low FOV camera setting..

I understand the confusion though; i should have rephrased that sentence

Thank you anyway for trying to help!

[Fredz]

Hi Fredz, we were talking about focal length values on the previous posts, so I meant that I'm interested in the FOCAL LENGTH value for the low FOV camera setting..

Ah sorry, I didn't remember the original question. You can find some info in the source code of psmoveapi but it doesn't say if it's for the low or the wide FOV : https://github.com/thp/psmoveapi/blob/m ... _tracker.c

Code: Select all

#define CAMERA_FOCAL_LENGTH 28.3	// focal lenght constant of the ps-eye camera in (degrees)
[...]
float cam_focal_length; // in (mm)
[...]
tracker->cam_focal_length = CAMERA_FOCAL_LENGTH;

[PatimPatam]

@Fredz

Hmm funny in that code they define a constant in degrees and then assign it to a var in mm. Probably just a comment mistake though, 28mm sounds about right.. but for the wide fov setting unfortunately..

Anyway don't worry, i'll find out one way or another! Thanks!

@Chriky

I share your pain with color led detection, did my own blob analysis as well, managed to get it quite stable now.. here's a couple suggestions you could try in case you haven't yet:

- try with different camera gain and exposure settings, i run mine very close to the lowest values possible (GAIN=0, EXPOSURE=15)

- control the brightness of your leds with the current intensity. Use a diferent serial circuit for each color and increase the intensity on the dim ones or decrease the rest to make them more uniform. The LM317T current regulators are great for this and cheap as chips.

- one thing i noticed with the leds i'm using is that the rgb values vary quite a bit with distance. This is a bit more complicated, but you can try to calculate the approximate distance first (easy with your fixed pattern) and then vary your color conditions or thresholds for each blob depending on that distance.

I haven't implemented this last bit myself yet, but i'm planning to in order to increase the detection range.

[Namielus]

chriky have you abandoned this project?

[Chriky]

I got to the stage where it was difficult to work on it in small pieces; I needed whole weekends minimum, which was impossible to fit in. I am hoping to do some more VR stuff after I get my Rift (and probably after Google Code Jam)

[Namielus]

So its a time issue rather than money issue then?