N95 camera distortion part 2

Could be wrong.
Actually what caused my presumed mistake was this research on the Nokia N95 camera calibration
N95 Distortion Evaluation

They got huge distortion estimation – like 3% of the screen width. I don’t detect anything like this in the 320×240 viewfinder. My upper bound is like .7% (though I didn’t do complete calibration, just did some rectangles shots).

N95 camera distortion ?

I’m making full screen version of my AR Tower Defense game, and stumble at some strange problem. It seems to me the camera distortion is absent if I’m getting scaled down viewfinder image (or may be it’s just cropped). If I’m getting full viewfinder image it looks like some camera distortion creep in – seems markers getting distorted. I may have to calibrate camera, and that would create portability problem. The game will work only for N95 camera…

Samsung INNOV8 and AR

Samsung INNOV8 (Samsung GT-i8510) seems to be the best Symbian device for AR so far.(And for 3d gaming too BTW)
Camera: 120 fps in QVGA vs 30fps Nokia N95
CPU: 450 Mhz ARM11 + PowerVR 3d accelerator vs 330 Mhz ARM11 + PowerVR 3d accelerator for Nokia N95
Sound: 3D surround sound effect by DNSe 2.0 technology vs ??? for Nokia N95
GPS : both, but INNOV8 being Symbian 9.3 allow self-signing for developer who want to use GPS.
For Nokia N95 use GPS require complete Symbian Signing.

Overall pretty impressive device, with 3d accelerator, 50% faster CPU and a lot faster camera.
Mobiles on other platforms often have better CPU (533-800 Mhz), but usually lack 3d accelerator and/or have bigger screen resolution. Bigger resolution would either eat more CPU power or would force developer make effective screen window smaller.

PS There are reports that actual INNOV8 CPU speed is 330mhz

Marker-based AR application and games could be viable.

Blair MacIntyre written in his blog that he consider markers for AR to be a “prop”: “As soon as your start requiring props, that makes the games less portable … a conundrum to be sure.” Whoever it could be that marker-based AR games/app are not actually so bad.
Edge report that “79 percent [of those polled say] they use their portable device in-home, far more than any other location.”
The interesting thing that in-door marker-based tracking have no disadvantages comparing with markerless. Markers could be easily placed on any flat surface, or taped/pinned to vertical surface. Most of surfaces are at the right angles, so one can even make easy to track non-planar multimaker arrangement – like putting markers both on the table and the adjacent wall. In some cases, like on smooth uniform surfaces only marker-based tracker can actually work, markerless would just fail there. Of cause markereless tracking is more scientifically impressive and fun to develop, but marker-based works right now )))
From business point of view marker based apps and games can provide interesting opportunity – non-intrusive mobile advertising. Instead of pushing ad into the face of the user, on the screen, occupying precious screen real estate, ad or logo can be put on the marker itself. That way it remain both highly visible and non-annoying. Different ad bearing markers could be keyed to different content units, thus providing ads rotation.

Augmented Reality Tower Defense game is up for download

It’s here. In this demo I’ve used the same multimarker tracker I’ve used in last demos. 3d position of the camera calculated form markers used for placing tower. Third dimension used to control tower height
If you don’t know what Tower Defense is, here is wiki entry

Symbian AR: N95 still the king of the hill

New Nokia 5800 is looking good, but if rumored specs are correct it still can not beat N95 (or N82 and other OMAP 2420 phones)
Nokia 5800 is rumored having Freescale i.MX31 369 Mhz CPU, vs Nokia N95 TI OMAP 2420 330 Mhz CPU.
Both have floating point (VFP), but only N95 have 3d accelerator, which would more then compensate for slightly slower CPU.
Nokia 5800 high resolution screen actually play against it in 3d applications. High resolution screen require more processing power for 3d rendering, which combined with lack of 3d acceleration can slow down 5800 considerably.
Touchscreen is interesting thing by itself, but not easy to integrate with AR from usability point of view. So N95 still win. IMHO.

The Greater Augmented Reality

That was in response to excellent Tim post.

I completely agree that “self” becoming progressively more fuzzy concept. For example “Self” concept usually include one’s memories. But what if some of my memories are stored outside of me, and my brain storing only search keys to them ? Yes, I mean google. Google already works as “Augmented Memory”. Ironically I’m developing mobile augmented reality apps, using google as augmented reality memory in weirdly recursive loop…

Nintendo DSi and Augmented Reality

It seems Nintendo DSi could be a good platform for AR. It already include video-processing AR-like tools, which have color or texture segmentation and some feature detection, though feature detection doesn’t seem especially stable, as could be seen on this video.
Most interesting thing though is DSi new CPU. bunniestudios.com report: “I can tell the CPU is substantially beefed up (consistent with reports of the DSi battery life being shorter than the DS-lite, despite having similar battery capacities…)” If DSi really have around 300Mhz CPU you can expect about the same performance as Nokia N95 for image registration. So my AR demos would have little quality loss for DSi comparing with N95. Lack of 3d accelerator for DSi would hurt of cause.

PS Nintendo DSi and Augmented Reality revisited

To filter or not to filter

Time-domain smoothing filter is somehow controversial question in image registration. The video is inherently smooth, necessity of time-domain smoothing is usually sign of instability of image registration algorithm. Taking average for several frames is especially notorious. Ideally good, stable algorithm don’t need any smoothing in time-domain. However what to do if there is small jittering in otherwise stable and fast real-time algorithm. Just couple of pixel in amplitude, it’s not noticeable on the big virtual objects, but on small, especially stretched or linear objects it’s quite noticeable. I have found that at least in my case, simple, several frames, smoothing filter remove jittering and is not causing any visible alignment artifacts.

Marker based augmented reality on CNN

Net is buzzing about CNN hologram at election night. What this tech really is the augmented reality + 3D imaging technology. CNN had 35 cameras in a ring at Obama headquarter to capture Jessica Yellin image from any possible 2D direction, with angle delta small enough not to be noticeable. After correct frame was picked up it was inserted into main videoframe. It seems to me the red circle on the floor was used for positioning of the image, as fiduciary marker. If you pay attention to the last part of the video you can see that Jessica “float” a little over the circle. That is because the circle is not the best choice for the marker. Square is probably the best fiduciary marker possible. It have distinctive corner, which could have subpixel accuracy, because they calculated as intersection of linearized square sides, and it produce third dimension axis easily, as cross product of cross products its sides.