Meant to be Seen

We discovered this one by accident.  What started as a satellite view got enhanced to a street level view.  Is the next step for Google Maps a stereoscopic 3D option?  You decide:



At this time, Google Maps only supports 3D in anaglyph (red/blue glasses).  However, looking at the images with the naked eye, it's unclear if they are true 3D images, or if they are 2D images which are just offset from each other.  For example, look at he painted lines on the road.  In practice, the lines should angle outward or apart from each other as they go deeper into the scene.  Instead, they are nearly parallel.  The objects in the far distance should have at least some separation, but there is none to be seen.

The earlies blog we could find talking about this updated was written on April 1st, so maybe it was intended as a joke?

To get the 3D options, go to "street view", and then click on the anaglyph glasses logo on the left hand side.  Share your results and findings in the comments section below.
Read full article...

I see the van is nearing depth from stereoscopic 3d but the total image shown is mostly flat.

it's a fake stereo...really bad

sort of, I think they generated it with the depth information of the laser distance scanners hat are mounted beneath the cammeras.

What started as a satellite view got enhanced to a street level view.

A picture taken 100's kilometres away will not have any stereoscopic 3d with a standard distance between the lenses.
Actually you start loose stereoscopic depth when the distance goes above 2-300 metres or even less.
At "spacedistances" then cameralense needs to be very far away from each other to experience any 3d at all. Besides: if going a "dual satellite" stereocamera i can imagine encountering other problems than just the cost . Doing a satellite "cha-cha" ie pictures taken with a couple of seconds between will give a more stereoscopic view but then moving objects won't be captured properly.

I think it's kinda nice that they've even thought about trying this. It looks like they're making use of their street/building info (the same info used for the mouse cursor 'collision' and the nice parallaxy transitions). So obviously the depth info will be very coarse... but I still like it.

However, their actual stereo rendering implementation is wrong. They haven't applied any kind of horizontal projection offset / image shift. This is one of the two classic mistakes that programmers seem to make when they try to create a 3D mode (the other being when they add toe-in to compensate for their error).

What can we do to stop this happening, again and again? I reckon we need a very clear, attractive tutorial explaining how to program stereo rendering correctly. Does anyone know of such a tutorial? If I ever get round to making one, I'll be sure to let you know :)

It it something serious ? i thought it was an april fools ?

What can we do to stop this happening, again and again? I reckon we need a very clear, attractive tutorial explaining how to program stereo rendering correctly. Does anyone know of such a tutorial?

This one is the most cited reference when talking about correct frustum calculations :
http://local.wasp.uwa.edu.au/~pbourke/m ... reorender/" onclick="window.open(this.href);return false;
And this one can be useful for filming too :
http://local.wasp.uwa.edu.au/~pbourke/m ... ereo_film/" onclick="window.open(this.href);return false;

I thought it was an April's fool joke too when I saw this on 3DGuy.tv, even if it was dated from April 2nd. Maybe they did just have their maths completely wrong, but I doubt it, we are talking about Google here.

No it's no "building based maths" what happens here would be the same if you were to go to google street view normally, pan the view horizontally a bit and create a 3d image after that - that's why what you end up with is, well, crap, since you have just one eye/zero separation - what makes this work even slightly is the fish-eyeness of the eye

Here's a couple of grabs that to me demonstrate the street/building shapes being used for the stereo. All I've changed is the image separation.

(Click the thumbnails for bigger versions)

Have you guys actually, you know, put on anaglyph glasses and tried it? Because if you did you would see it was 3D. Now even though there is real depth, they are not stereoscopic images. What it is doing it using some sort of underlying polygon wireframe to warp the image a bit. So I think they model blocks for the buildings but they don't model the cars or lampposts, etc. However it does look decent for what it is and is an interesting feature. Will I use it, probably not, but its still cool.

The first image posted here looks like crap and it's not even good as being a conversion and honestly i didn't even bother trying it out.
I have to say that the last two looks really great! It' no real stereoscopic image and not entirely correct in every aspects but the algorithm seems to work way better with higher separation!