EEVblog #734 - Giroptic 360cam Kickstarter Prototype

[EEVblog]

Dave takes a quick look at the first run Kickstarter developer prototype 360 degree waterproof HD camera from Giroptic
It produces real time 360deg 30fps MP4 video with spatial audio in
360deg video on youtube is here:
Kickstarter Project: https://www.kickstarter.com/projects/giroptic/the-worlds-first-full-hd-360-camera
http://www.giroptic.com/
http://360.tv/

[Tek_TDS220]

Cool product, but the sound reminds me of the old Myst game where you open one of the time travel books: "Bring me.....  the.....  blue pages...    the.....  blue....  pages"

[hikariuk]

That audio is bad enough that you really have to mute it or it drives you nuts.

I'm actually kind of impressed by the way YouTube handles 360 video.

[Fungus]

I'm actually kind of impressed by the way YouTube handles 360 video.

?? I'm not.

Edit: Oh, wait you have to use Google Chrome to view it....

PS: Is that an arcade machine over in the corner behind Dave II? What game? The video is too low res. to see.

[T3sl4co1l]

Wow, the 360 player must be ridiculously processor intense... I've got plenty of graphics in this machine but I can't play anything beyond 480p in real time!  Lots of optimization to go, I suppose.

Tim

[firewalker]

This is what I get under Firefox on GNU/Linux. Both on HTML-5 and Flash players.



Is it a Chrome only thing?

Alexander.

[Fungus]

This is what I get under Firefox on GNU/Linux. Both on HTML-5 and Flash players.

Yep. That's what I got as well.

Is it a Chrome only thing?

Looks like it.

[free_electron]

i like the idea, the case design is very clever, like the led matrix . this thing is full of clever design bits, excellent.  but... ,pardon my french, the video is CRAP.

Now, that being said : from the flaring around the lights in your ceiling : either the lenses have greasy fingerprints on em , or there is a protective film on em you forgot to remove. or the lenses are out of focus.

If that ain't then they got some mayor work to do on the camera optics. i would expect at least 'gopro' quality video ( and that's not really high quality )

[Muxr]

I dunno about the clever case design. It would have been smarter if they put the heatsink at the top and away from the battery.

[German_EE]

First comment from my girlfriend, "it looks like a frog, why don't they make it in green"?

[bitwelder]

Dave could have let it run for a while and the open it and measure the internal temperatures (and compare with the external bottom part).

[Unixon]

How about doing an X-ray scan of that thing?

[Grapsus]

Looks promising but a lot of work remains on image quality and audio.

I'm not sure that stitching is such a huge processing deal, as long as all the sensors and lenses remain at a fixed position. Once the transformation has been computed for one set of images it could be applied to all subsequent frames and it becomes just a matter of moving the data around. Still it's not ideal, in the demo video we can see a stitching line when Dave moves around. This suggests that they don't even blend images from several sensors together, it's just three vertical stripes with data from each sensor.

Edit: Don't expect the 3D view on Youtube with anything else than Chrome. Even if it was possible with another browser why would Google bother implementing it ? It's like the Maps application which became unusable on any other browser after an update, another gentle pressure on the users to install their stuff.

[Agent24]

Cool product, but the sound reminds me of the old Myst game where you open one of the time travel books: "Bring me.....  the.....  blue pages...    the.....  blue....  pages"

Ha, it totally does! Although I think the audio in Myst was even more distorted.

First comment from my girlfriend, "it looks like a frog, why don't they make it in green"?

First thing I thought of was one of these: http://ocean.si.edu/ocean-photos/grimpoteuthis-dumbo-octopus

[EEVblog]

I'm not sure that stitching is such a huge processing deal, as long as all the sensors and lenses remain at a fixed position. Once the transformation has been computed for one set of images it could be applied to all subsequent frames and it becomes just a matter of moving the data around.

Ah, yes, true. I'm thinking of regular stitching of images that require figuring out how the images have to best blend together because the angles and field of view etc are always different. In this case everything is fixed and known, so yes, should be much easier and quicker.

[EEVblog]

Now, that being said : from the flaring around the lights in your ceiling : either the lenses have greasy fingerprints on em , or there is a protective film on em you forgot to remove. or the lenses are out of focus.

I cleaned the lenses before shooting. Apparently they are clip-on lenses some how.
Yes, the video quality is pretty crap, they have a long way to go before Joe Public will want to use this.

[mux]

The primary reason why this is processor intensive is much simpler: there are simply NO fixed-function video codecs for this resolution and aspect ratio. The stitching isn't a problem; encoding is. You're definitely going to need some fancy media processor or FPGA to do that.

[Grapsus]

The primary reason why this is processor intensive is much simpler: there are simply NO fixed-function video codecs for this resolution and aspect ratio. The stitching isn't a problem; encoding is. You're definitely going to need some fancy media processor or FPGA to do that.

That's a good point. Maybe they have been too ambitious here by trying to get 3*HD resolution ? Another possibility would have been to just use a small FPGA or CPU for the stitching and down-scaling everything to a standard resolution so that an integrated low-power encoding chip could do the encoding ?

[EEVblog]

That's a good point. Maybe they have been too ambitious here by trying to get 3*HD resolution ?

That would explain the low resolution and quality of the video, they just ran out of oomph on their processor.

And they just admitted on Twitter that audio issue is firmware. So they either:
a) Deliberately shipped known faulty firmware to developers
b) Didn't test the audio at all
 

[Fungus]

The stitching isn't a problem;

I agree. The stitching is a red herring because it's the same for every single frame of video. You only need to do the image processing once when they hit record then you can use the exact same distortion map for the rest of the recording.

(and the stitching didn't look great to me...there was an awful transition band when Dave walked around it)

[tchicago]

The primary reason why this is processor intensive is much simpler: there are simply NO fixed-function video codecs for this resolution and aspect ratio. The stitching isn't a problem; encoding is. You're definitely going to need some fancy media processor or FPGA to do that.

Why can't they encode three streams in each camera's native resolution using readily available fixed function hardware codecs, and then do stitching and reencoding on PC? That would save a lot of energy and extend battery life.

[Fungus]

The primary reason why this is processor intensive is much simpler: there are simply NO fixed-function video codecs for this resolution and aspect ratio. The stitching isn't a problem; encoding is. You're definitely going to need some fancy media processor or FPGA to do that.

Why can't they encode three streams in each camera's native resolution using readily available fixed function hardware codecs, and then do stitching and reencoding on PC? That would save a lot of energy and extend battery life.

There are FPGA-based codecs out there that should be configurable for frame size.

eg. This one says it can do up to 4096x4096 on a single FPGA:

http://www.a2etechnologies.com/products.html

I imagine a bit of tweaking could make it do 8192x2048 or whatever...

(found via google, I don't know anything else about it)

[bktemp]

A teardown would be interesting.
The processor/FPGA needs 3 camera inputs. Since 3 cameras is unusal for a hand held device, I doubt there are many small processors/GPUs/SOCs capable of handling 3x 1080p streams at 30fps, so is it most likely a FPGA doing all the low level acquisition stuff such as reading the 3 synchronized video streams and stitching them together.

[Grapsus]

Why can't they encode three streams in each camera's native resolution using readily available fixed function hardware codecs, and then do stitching and reencoding on PC? That would save a lot of energy and extend battery life.

That's because they're looking for a "frictionless user experience"  . The thing has WiFi, so they want the users to be able to press a button and boom, the 360° video goes on youtube. It's a noble goal as long as it doesn't sacrifice quality...

[firewalker]

This gadget doesn't stream 360 in real time?

Alexander.