sony actioncam

[2EyeGuy]

In the duty free at Sydney airport I saw Sony actioncam and gopro both 359 dollars each. I can't aford it though, especially if I need 2, so don't expect rift videos of my holiday to Europe. Sorry. The specs look really cool for the rift though.

[Namielus]

New gopros just got released, even tho the hero 2 is not even a year old , the gopro hero 3 is out with a beefy performance increase.
It has however, less distortion which would not be ideal - the good thing is that people are going to be selling cheap second hand gopro hd hero 1 and 2's and ive seen it for as low as 100 dollars each.

[mAchiNE]

New gopros just got released, even tho the hero 2 is not even a year old , the gopro hero 3 is out with a beefy performance increase.
It has however, less distortion which would not be ideal - the good thing is that people are going to be selling cheap second hand gopro hd hero 1 and 2's and ive seen it for as low as 100 dollars each.

Actually the GoPro HD Hero 3 has Ultra wide, Medium, and Narrow FOV settings at 1080p which if the same as HD Hero 2 is 180deg, 110deg, and 90deg respectively (from memory) however the HD Hero 3 does not appear to be compatible with the 3D Hero accessory so the best choice is still the HD Hero 2 if the fisheye distortion of the higher FOV settings on the camera is a good match for the Rift but I can't be sure that it is untill I get my Rift and try it out (my Brother has a 3D Hero setup with HD Hero 2 cameras and I have some footage of him flying a Helicopter at sea that I can test it out on), Although the 90deg setting (the one that looks more normal) is probably a better match for the FOV of the Rift if you want things too look right but the image will need to be pre distorted before watching on the Rift as I understand it.

[Namielus]

You can have a wide fov with more or less distortion. And the Gopro Hero 3 says it has less distortion but they wont say how much less. The barrel distortion of the gopro hero HD matches the pincushion distortion of the rift optics.
Therefore the result you see is a perfectly flat image. The gopro 3D housing has the lenses way too close and wont work.

It needs a custom IPD. 65mm from center of lens to center of lens. Look at my gopro HD thread, there is a perfect video there to see the correct distortion and IPD for rift content (contribution by BOLL)

[mAchiNE]

Ahh I see, I just assumed the distortion was linked to the FOV, I didn't realise they had somehow improved it, it is a shame about the 3D Hero lens spacing, their prototype 3D case had better separation, I can see why they changed it (due to how its normally mounted by their target market with something fairly close in view). It's good to know the distortion is a good match for the rift though, might have to have a play around with it when I get my rift

[Namielus]

You know your stuff! I have measured the pictures of the prototype (if thats whats in the pictures I found rather than third party product) and its ironically extremely close to 65mm.

I tried contacting them to get my hands on it, but they proved to be one of those companies that is very hard to communicate with.
The vertical cameras 65mm distance configuration however, will let you put an individual housing on each camera.
The built in wireless in hero 3 will let you sync the cameras without connecting them anyway.

The only question that remains is how different is the distortion. Unfortunately they wont even tell me that.
I hate it when companies keep basic information about their products a secret.

You put the cameras vertically because of the aspect ratio each eye will have in sbs-mode on the rift.
The resolution in that configuration is good enough up to 240 frames per second on the hero 3 (850x480@240hz per camera)

[Skaven252]

The barrel distortion of the gopro hero HD matches the pincushion distortion of the rift optics.

Does it? Hmm, I wonder how it matches with the VIO POV.HD. I've been trying to shoot 3D with a pair of those, head mounted.

The built in wireless in hero 3 will let you sync the cameras without connecting them anyway.

Have GoPro confirmed this? Are they actually genlocked wirelessly, or does that just mean they can be started and stopped simultaneously? The sensors can still be off phase if it's only a case of the latter.

[Namielus]

A gopro rep told me you can start an stop quite a lot of gopros with one remote, yes. However, I have also been told that the wireless features either the remote or smartphone app can replace the gopro 3D "sync cable".

I have not confirmed the latter yet.

[Skaven252]

I have also been told that the wireless features either the remote or smartphone app can replace the gopro 3D "sync cable".

Alas, not even the sync cable actually genlocks the cameras. It only estabilishes a master/slave control relationship between them.

See this review, this guy tested it thoroughly:
http://www.chrisevans3d.com/pub_blog/?p=604

I can only hope that they'll introduce wireless genlocking in Hero3.

[EdZ]

Wireless genlocking

I'm not sure it's even possible to do such a thing without it being hideously expensive (maybe some sort of global sync chirp to reduce multipath errors, with a twinned time-of-flight signal for each remote clock box for distance offset?). If GoPro were willing to put an accurate clock into the HDHero, you could have them touch contacts to initially sync clocks, then remain reasonably in sync for a good hour or two if continually powered after separation.

[Namielus]

How good/bad is the sync when its basically just recieving the start signal as closely together as possible?
How much of that can be fixed in post?

[EdZ]

How much of that can be fixed in post?

Essentially it can't. Fixing anything more or less than an exactly multiple of 1/framerate is very, very difficult, and best described as an ugly hack: interpolate to framerate = 1/offset (e.g. 4ms timing offset = 250fps), then pick temporally matching synthetic frames and deal with the interpolation artefacts. Unless you have a display that can produce these incredibly high framerates so the offsets can just be displayed, you either deal with quality loss, or hope the timing offset is so small that nobody notices if you play them unmodified.
An analogous problem is using multiple cameras for live TV: all the cameras need to provide their frames to the central mixer within a few nanoseconds of each other or horrible frame losses will occur, so the cameras need to be 'genlocked', or fed an external timing pulse common to all cameras and AV systems. The same is usually done for recorded TV and movies to make editing a lot easier (mainly timing for multi-shot SFX and sound sync).

[Namielus]

The new hero3 can film at 720p/120hz, lets say you just move one of the timelines so the frames match to the closest possible in the other eye but they are slightly offset in time;
Lets assume maximum 1/120 second offset) and probably less since the wifi remote will probably start them pretty much at the same time, then the video is encoded to a 60hz file.

How noticeable is it if one eye is offset in time like that for the frames to match exactly? If these questions make no sense I am not sure I understand the challenges.
I assumed the goal is to have both frames being drawn at the same time and therefore should also match in time.

[EdZ]

The new hero3 can film at 720p/120hz, lets say you just move one of the timelines so the frames match to the closest possible in the other eye but they are slightly offset in time;
Lets assume maximum 1/120 second offset) and probably less since the wifi remote will probably start them pretty much at the same time, then the video is encoded to a 60hz file.

How noticeable is it if one eye is offset in time like that for the frames to match exactly? If these questions make no sense I am not sure I understand the challenges.
I assumed the goal is to have both frames being drawn at the same time and therefore should also match in time.

If you're filming at 120Hz, your max offset will be a hair over 2ms (half of 1000ms/120fps), assuming you have a nice marker in your footage to align to the closest frame. Even if you drop down to 60fps, you'll still have the same offset. There's a good chance you won't notice this small an error except with fast camera/object motion. With all the extra stereo context you're getting from the RIFT, hopefully the convergence shift caused by frame mismatches will be 'overidden' by your brain so you don't perceive the problem. It's really something you'd need to try out in practice to know for sure.

[Skaven252]

Also, not only the cameras may gradually shift out of sync due to the tiniest difference in clock speed, cameras with CMOS sensors may also drop frames. And this happens very inconsistently between the cameras. Even if you started them simultaneously, and their clocks ran at a reasonably matched speed, the dropped frames may still mess up the sync.

Wireless genlocking

I'm not sure it's even possible to do such a thing without it being hideously expensive (maybe some sort of global sync chirp to reduce multipath errors, with a twinned time-of-flight signal for each remote clock box for distance offset?).

I'm not familiar with the tech required, so could you explain what's so difficult about synchronizing two sensors wirelessly, but not via a wire?

The LANC system makes genlocking possible, does it not? And it operates at a low data rate. I'm assuming that commercial 3D camcorders that have twin sensors are also in sync, and this is done onboard over a wire connection (presumably).

Radio waves travel at the speed of light. I don't get it. Why is it so difficult to base the synchronization on radio waves?

[EdZ]

Radio waves travel at the speed of light. I don't get it. Why is it so difficult to base the synchronization on radio waves?

Path length. You'd need to compensate for this when using a wireless clocking system, whereas when wired you can either just measure the wire length and add any extra needed to ensure all endpointss are the same distance from the clock gen, or you can add a fixed delay to each line to bring them all in sync. Once you've plugged the wire into the camera, you can wander around without changing the delay. With a wireless system, the delay would be different for every receiver, and moving around would change the delay.

[Skaven252]

I recall the human brain is actually surprisingly forgiving about sync disparity - as long as each eye get its own separate picture. If there's ghosting or crosstalk, you 'll see double images (fast moving parts of the picture driting apart) if the sensors are off sync, and that can get jarring.

BTW, speaking of 3D action cams, the Camsports Fusion 3D looks quite interesting. They're claiming that the two sensors are in sync - but it records on two separate memory cards. I wonder what the reality of the device will be.

[Skaven252]

Path length. You'd need to compensate for this when using a wireless clocking system, whereas when wired you can either just measure the wire length and add any extra needed to ensure all endpointss are the same distance from the clock gen, or you can add a fixed delay to each line to bring them all in sync. Once you've plugged the wire into the camera, you can wander around without changing the delay. With a wireless system, the delay would be different for every receiver, and moving around would change the delay.

Thanks, that explains it then. I suppose the LANC sync cables are only available in standard lengths then?

The changing distance between the cameras should not be a problem for stereoscopic video, because the camera pair is usually mounted at a fixed distance apart. Unless you make a rig with adjustable convergence... hmm yeah, there's a moot point. If you adjust the convergence by rotating the sensors instead of shifting them horizontally, the distance shouldn't change that much though.

I was thinking that if the signal really travels at the speed of light, shouldn't it be practically instantaneous regardless of whether the distance between the source and the destination changes or not? It's not like the cameras are going to be more than a few meters apart at most, and the most commonly used distance would be 65 mm (the average human IPD).

[Skaven252]

I guess the thing I'm getting after is this: with the short distances between the cameras involved in stereoscopic video, if you send a radio 'ping' from one camera to the other to sync the frames, how long does it take to reach?

If light / radio waves travel at 300 000 km/s, shouldn't the delay between two cameras that are 65 mm apart be like 1/46 121 916 61th of a second? Seems like a negligible difference, even if the distance between the cameras changes. Even if they are several meters apart, the lag is there, but it should be negligible, right?

What remains then, is the inconsistencies between identical cameras' internal electronics. How big are those inconsistencies usually, if the cameras are manufactured the same way? Can the camera measure them and take them into account? The goal would be just to get the CMOS sensors be read at the same time, or at least "close enough".

Are video formats generally aware of dropped frames? So if a frame is dropped, the previous frame's duration would be set to 2 frames (or the previous frame would be duplicated)? This way the videos could be kept in sync, even if frames are dropped inconsistently between the two cameras. The biggest problem is less-than-a-frame off sync. If the sync is off an exact number of frames, it's easy to fix in post. Until we get those random dropped frames between the two cameras, that is.

So what I'm shooting for isn't "acting exactly in sync like North Korean group dancers" but rather "check regularly we're more or less on the same word on the same page". Would that be more feasible over WiFi?

[EdZ]

You also need to encode and decode the radio (or IR) signals. With a sync cable, you can pretty much just hook one oscillator straight up to another with 0 delay and minimal cost, whereas if you need to read the clock signal, encode it, amplify it, discriminate it, decode it, and output it again, that's an additional delay you need to compensate for, and a lot of extra components to add (as well as a method to 'pair' the cameras).

So what I'm shooting for isn't "acting exactly in sync like North Korean group dancers"

Unfortunately, for video you need to be much better than even the best DPRK dancer. Remember that errors can compound as you video runs, so a 1μs (1 microsecond, or 1 millionth of a second) error range per frame, with a 60fps 15 minute video for example, you could be up to 54ms out by the end in a worst case scenario, or 3 frames! On average things won't be nearly this bad, and having the clock signal be unidirectional helps a lot to prevent errors building up over time, but with something like the GoPro where the oscillator is aimed at being small and cheap rather than sub-μs accurate, you want to minimise error every way you can to prevent awkward problems cropping up later.

[Skaven252]

I realize errors accumulate over time... but could you do the correction periodically? Like, send a ping once per second, or 5 times per second, or whatevs, to check the sync. If it's off, make the camera compensate by dropping or duplicating frames, and/or by resetting the sensor phase. Constant, repeated "shepherd" correction instead of getting everything perfect from start to ensure a smooth run.

Would that be a too simplistic solution? At least to me (I'm no expert) it sounds like it could work even over really long takes, since the correction is done continuously. If they are drifting apart, they get promptly slapped back to their senses. The videos don't need to be perfectly and clock-accurately real time, it's enough that they are wobbly and off sync the same way in relation to each other.

[EdZ]

Dropping or duplicating frames is what you absolutely want to avoid. The goal with camera synchronisation is for each frame to be captured at as close to exactly the same time as possible. This time difference is pretty much always less than the time of a single frame, so dropping or duplicating frames will only make things worse anyway.

[2EyeGuy]

I recall the human brain is actually surprisingly forgiving about sync disparity

If two eyes of a 3D video are out of sync, you can fix it in wetware by holding a dark filter over whichever eye is too far ahead. Presenting an image to each eye at exactly the same time isn't the same as presenting each image at the same time to the part of your brain that handles stereoscopic depth perception. Your brain's separate processing paths creates it's own sync disparity when one eye needs more analysis time than the other because the image is darker.

[Nick3DvB]

Isn't that the Pulfrich effect? http://en.wikipedia.org/wiki/Pulfrich_effect

I found my old Nuoptix glasses the other day, it took me a while to work out they weren't some weird type of anaglyphs, doh!

[Namielus]

I laughed so hard when I saw black ops 2 for ps3 had 3d anaglyph mode.
I wonder how many people are using that

[Nick3DvB]

But I'm sure the 3 guys that do use it are very happy they induced it! LOL I had to use some on my XP box the other day, I forgot how bad the colour persistence is, I went to watch some TV afterwards and thought someone had been playing with the image presets again.

[2EyeGuy]

If you put the glasses on backwards for a short time, it will fix the white-balance in each eye.