New Augmented Reality Demo for Gizmondo

[Zaphod]

that's why i like the Gizmondo quite a lot at the moment. this is the very first time i got or saw something like this to work at a really high-fidelity frame rate (constantly above 25 fps).

Trouble with the Gizmondo is that it's going the way of the Dodo. Here's a video of similar stuff that you've shown running on last gen Series 60 (papers available as well). Certainly less impressive graphics wise than your demo (but if Nokia get their way) probably a more viable platform going forward.

Edit: Didn't notice that Lazy8s had already linked that. Silly me...

 

[Zaphod]

It was based on a modified version of AR Toolkit Plus from memory. ARTK+ is basically a modification of the standard ar toolkit, adding a few (desperately needed) features, and supporting fixed point processing.

FWIW: Here's an earlier S60 port of the ARToolkit from the same guys. I think that is the first smartphone AR-implementaion I read about.

 

[hidefguy]

Trouble with the Gizmondo is that it's going the way of the Dodo. ... Certainly less impressive graphics wise than your demo (but if Nokia get their way) probably a more viable platform going forward.

Yes, in fact it's already "gone", apart from Ebay and other private purchases, since Gizmondo Europe went bankrupt in January. I agree totally that currently, the best hardware bases are in Symbian - e.g. the N93, which we've got on order specifically for this purpose, but Windows mobile-based devices will be appearing within the next six months with more capable GPU's than the GF4500 in the Gizmondo.

 

[drgoldie]

Overall to be totally honest, the AR toolkit is pretty bad. The code is a shocker, and it's tracking performance can be down right terrible. It is difficult to setup as well. But at least they are finally investing some proper money and time into getting it into a more commercially sound state, however licensing the toolkit is still a struggle (to be kind).There are other alternatives out there. There is the MXR toolkit, which is *much* smarter than the ar toolkit, and has *far* better tracking performance, however it's code is probably even worse, is bloated, is no longer updated, is buggy, and has performance issues.

The ARToolKit source code is not beautiful. ARToolKitPlus is a little bit more cleaned up, but also not perfect. I don't know about the tracker of the MXR toolkit that much, but ARTag is to my knowledge the best marker tracking library at the moment (closed source and commercial though and no PDA/phone versions...).

FWIW: Here's an earlier S60 port of the ARToolkit from the same guys. I think that is the first smartphone AR-implementaion I read about.

The Hitlab's Symbian version of ARToolKit is basically just a very old version of ARToolKitPlus with Symbian specific modifications.

I agree totally that currently, the best hardware bases are in Symbian - e.g. the N93, which we've got on order specifically for this purpose, but Windows mobile-based devices will be appearing within the next six months with more capable GPU's than the GF4500 in the Gizmondo.

Can you back that up? Which Symbian hardware is better than the best current Windows Mobile hardware? I always had the impression that WinCE hardware was more powerful (which is also because WinCE wastes more resources than symbian...)

Which devices with new nVidia chipsets are gonna be released? I know only about the HTC Trilogy which will have a GoForce 5500. Any other?

bye,
Daniel

 

[Lazy8s]

As a starting reference, the Mobile Data Visualization site that's linked in hidefguy's signature provides an overview of many of the different 3D portable devices for the market.

http://mobile.sdsc.edu/devices.html

 

[hidefguy]

Can you back that up? Which Symbian hardware is better than the best current Windows Mobile hardware? I always had the impression that WinCE hardware was more powerful (which is also because WinCE wastes more resources than symbian...)

Which devices with new nVidia chipsets are gonna be released? I know only about the HTC Trilogy which will have a GoForce 5500. Any other?

By that I meant handsets with hardware graphics acceleration. There are already several symbian-based devices with hardware graphics acceleration available (or soon available) on the market that can be programmed in GLES (N93, M600, P990, W950i - the OMAP 2420 in the N93 appears to represent an especially powerful development platform), while there are (as far as I know) currently two accelerated devices with CE/WM5 that can be programmed (the X50/51v and the Gizmondo, both of which are EOL'd or defunct). Unfortunately, the Samsung SGH-i310 running WM5 with the GF4800 that's just started shipping has no hardware 3D drivers (neither GLES or D3DM), but the HTC Forseer (called Modeo in the US) will be released sometime later this year, which uses WM5 and a GF5500, though it's unclear, again, whether this will ship with 3D drivers. As far as I'm aware, the Trilogy doesn't use a GF GPU. I've been told to expect more Windows mobile based devices with programmable 3D drivers in the first half of next year.

 

[drgoldie]

at the moment we strongly concentrate on WM devices, but maybe it is time to look out more for Symbian devices too. thanks for opening my eyes...

Daniel

 

[Graham]

I'll have a look into ARTag. Looks good.
I know I have seen that site before... Can't remember when.

The major difference between the MXR toolkit and the ARtoolkit is the pattern design. MXR patterns require a black border that is not connected to the internal pattern, and also require a notch in one edge.

The AR toolkit simply finds corner points in the thresholded image, then links them up, trying to find squares. The problem of course is the entire maker must be visible. MXR, on the other hand, tracks the outside edge or internal edge of the border, or both.
The notch means the orientation of the maker is already known. One major limitation of square markers is that the pattern cannot be rotated, or even reflected (AR toolkit takes into account markers visible in a mirror). This greatly reduces the variability of the markers and also slows it down a lot. Simply loops the recognition code 8 times.
The image comparison code in the AR toolkit is also quite poor, it's 'confidence' rating for matching markers is simply a dot product of the corresponding pixels. The default minimum confidence is 0.5 if memory serves :/ MXR toolkit is smarter here too. Both do allow for full colour markers at least.

That said the mxr is just as difficult to work with.

 

[drgoldie]

I'll have a look into ARTag. Looks good.
I know I have seen that site before... Can't remember when.

ARTag is great. I believe it is currently the most advanced "square-marker" tracking system. unfortunately it is closed source and there is no PDA/Phone version.

The AR toolkit simply finds corner points in the thresholded image, then links them up, trying to find squares.

Actually it is the other way around. ARToolKit first finds the squares, then extracts the edges and uses those to calculate the corners. It then uses the edges for the first (initial guess) and the corners for the 2nd (refinement) pose estimation.

One major limitation of square markers is that the pattern cannot be rotated, or even reflected (AR toolkit takes into account markers visible in a mirror). This greatly reduces the variability of the markers and also slows it down a lot. Simply loops the recognition code 8 times.

This is only true for the standard ARToolKit template patterns. It is only slow when using many markers (e.g. 30+). ARToolKitPlus' id-based markers do not share this problem. One can have 4096 markers at a higher speed than a single template marker. The rotation is simply checked by ruling out bit-masks that are not valid...

Daniel

 

[Lazy8s]

A new AR application that lets construction and utility workers see the layout of underground gas, power, and cable lines of an area in the city is being prototyped.

http://studierstube.icg.tu-graz.ac.at/handheld_ar/vidente.php