When small interaxials just don't cut it! - Page 24

I used to wonder about this drift issue. Does it really matter? Is the way you all are measuring it accurate? How do you know it is accurate? I prefer to look at the end results. If you pair the left and right camera clips using sound sync and a clap board as the slate and then set the editor to quantize to frames ( what is more important ) a sound sync que at the beginning and at the end of a 10-15 minute clip has virtually no drift at all. Plus, the number of frames on each left and right eye clip is exactly the same number between the two sound slates recorded. I have never experienced two clips, that drift more than a frame, ever, and the sound match is so close it doesn't matter. What matters is that the two clips match frame start at the beginning of the clip and at the end, which mine always do. In addition the 3D pair quality is dead on by visual cue as well, meaning that a fast moving object in the shot ( the clap board ) has the same visual reference in time at both ends of the clip for both cameras once lined up.
My experience has also been that the reliability of exact frame match between the two cameras can be out by 1-3 frames on startup, therefore the record start point based on clap board sound and visual reference at the start of the recording is a good safety measure ( actually I say it is required ), however, once the two clips are properly aligned ( quantized to frames) the end of clip second slate with the clap board has been shown to be completely unnecessary, redundant at best. In otherwords, I never rely on the Lanc Shepherd to establish the beginning of my paired clips. I use it to start the two cameras recording and use the timing meter as a reference to show it is working. Then I slate the cameras with the clap board or even my hands clapping in front of both cameras. ( If I'm in a crowd I don't use the clapboard because it draws unnecessary attention. ) But I still slate the start of the clips. Once I line the two clips up in Vegas at the clapboard marker, I the slice both clips in front of the slate marker. Most of the time, the clips are dead on with the Lanc Shepherd start record, but about 5% of the time they are not.
I dare say that if your camcorders are drifting sync by a significant amount it is likely there is something wrong with them and they need to have the capacitors checked in the cameras sync circuits. And, yes, I've been there done that with my one DXC-637 cameras that would not genlock in a multi-camera shoot. When the caps get old, like 10 years or more, the camera's timing will drift. But this is a repair issue, not a design one. There is even a Sony Tech white paper on the topic. When repaired the same camera could be used in a genlock environment, could hold field accurate for a whole day on it's internal clock.

Well, I can compare the behaviour of my old TD10 with the new one. The old one had a wonderfull low drift - something about 0.016 ms/minute compared to the first new one. The second new one, compared also to the first new one, shifts with 0.2 ms/minute - 14x faster.

The funny thing is - you hear that when you startup the camcorders using the controller. Sometime they start really equal and sometimes there is a time difference.

No, there is no error in the camcorder. What I understood from the German Stereoforum.de is that a well-known phenomena, that different cameras show different shift speeds. That is why they are running pairing actions - testing different combinations of camcorders to find out the combinations that work best together.

Funny. Now I have a combination that is fine in terms of optical geometry, and also a combiation that is fine in terms of synchronisation. So I have to look for another TD10...

I have also seen the situation that the paired clips have different length - but I do not know if there is something missing at the beginning or at the end or if it is different all the time. I have seen the movement of a tree due to wind - and the synchronisation was not perfect - you see that.

Actually, you have no idea what is right or what is wrong. You've heard the old saying- I bought a watch and thought I know what time it was. So I bought a second watch to check on the first one. But it was different. Before I thought I knew what time it was, now with two watches I don't.

The point is, unless you are comparing to a time standard, called a bench mark, you don't know. All you have are different control points. You could buy a dozen camcorders and find two that based on your meter are close but other than wasting money, none of it really matters.
If you can match the two clips at a same point in time that is known ( a visual and audio moment), then run clip from there, it doesn't matter if you started one camera 5 minutes before the second. As long as your two cameras video is used between two control points ( two clap board references ) and the number of frames between the two are the same, you will be locked as good as necessary. You don't need a benchmark ( a time reference standard), just two control points. You can shoot 3D video with excellent results without the sync device. That just makes things simpler but doesn't ( IMO ) negate the necessity for the clap board control points of reference.

If you want trouble, just use your sync device and shoot with no sound, no visual cue. You may have seen a 0.5 ms time sync but if the clips are different by 17 frames because one started before the other, you won't have a clue what the line up should be.

The bigger issue is the optics alignment. Looks like you have that under control.

Given the firmware that tends to choose the shutterspeed as required in automatic modus, and that can be short speeds too, I also think that the pair of cameras should be in sync. The manual of the SteFraLanc says that a missync should not be higher then 4 milliseconds. And Werner Bloos who sells that units wrote me - similar to Frank - that he would not accept a missync larger then 1 ms.

And for Lanc Shepard one can read: "I think that as long as the shutter sync is less than about one tenth of the shutter speed, disparity of the stereo images related to shutter sync will not be noticeable. So if the camera selected shutter speed is 1/100, then a shutter sync displayed on the LANC Shepherd of 1/1000 or smaller (say 1/2000, 1/4000, etc) will result in images with no noticeable motion disparity." So they also talk about 1 ms.

http://www.ledametrix.com/lancshep/user.html

The funny thing is that I see the difference here within the 3 cameras. While the combination of the old camera/the first new has a drift of 0.016 ms/minute, the combination of the first new one/the second new one is about 14 times higher. Means, that it takes for the first combination 30-60 minutes before the missync reaches the 1 ms, while it is for the second combination something about 5 minutes only. The second issue is - the combination of the old TD10/the first new one starts up with a missync of 1-2 ms a lot of times, while the second combination tends to start up with 0,2 - 0,5 ms a lot of times.

And well, I am able to exchange the newest camera within 2 weeks free of charge. That is why I will check another camera and then have a look to what combination is the best for me.

Wolfgang- You were reading about the sync issues for STILL photography in the Lanc Shepherd paper. Here, I agree it is extremely important since you are having to make both cameras snap the same moment in time just ONCE! That paper says very little about video other than to how to start the cameras and that the controls will work similarly. With video the "sync" story changes. Each camera is designed shoot a series of frames that are contiguous ( adjacently one after the other in progression ) that are already precisely timed to 1/23.976 seconds long. This fact of video makes it possible to line the two clips up in Vegas timeline exactly when quantize to frames are on. Here you ignore the clip start point and follow the marker. You could use any method to place a marker on each recording. Many videographers choose to use a strobe flash for multicam shooting to place a timing marker but I, coming from a film background, use the traditional clap board slate. With professional video cameras, you could use drop frame time code for the lineup. The Vegas explanation uses the audio track and explains how to relate that to your video sync. It's no different than stated earlier.
I now understand why you and Frank are getting so confused over this sync issue, You are reading something that IS IMPORTANT to still photography and assuming it is equally important in video. It's not because we can line up the timing in the timeline. You don't get that opportunity in still photography because if the images are not in sync enough for the shutter speed, your match moment in time will not be the same. In video you just slide the clip to the next frame to achieve the exact match. In video you can't quantize to frames shorter than the frame rate.

Instead of reading how to sync still photography for stereo, you should study the section in Vegas Help file. In the index- type "Pairing" and then display the list of topics. A good read is how to edit paired clips. Select "Stereoscopic 3D Event Synchronization". Here is the proper explanation on how to edit paired video clips for proper sync. The discussion here does not confuse the reader with unnecessary topics that is important only to still stereo photography.

When Don refers to frame quantization with Vegas Pro, I assume he means aligning the two clips to the nearest project frame, rather than to any fraction of a project frame (coupled with some kind of motion interpolation algorithm). If I am wrong, may I ask where do you find the motion interpolation feature on Vegas Pro? And does it work well? EDIT: I now see that Don has stated in the post above that quantization is to the nearest frame.

I suspect that for a quiet contemplative video with very slow pans, and using Vegas Pro to render the 60i capture at only 24p, it would not surprise to get away with two 60i cameras capturing asynchronously. With a Vegas Pro project rate of 60p, you could time align the Left and Right clips to the nearest 1/60th second. The worst case "phase discrepancy", assuming negligible relative drift for the Left and Right clips, would be half that, or 1/120th sec (8.3mS).

If you were using two cameras capturing at only 24p, the worst case phase disrepancy if using a project rate of 24p and time aligning to the nearest frame would be 1/48th sec, or 20.8mS, rather more significant.


I've done an experiment with my own sensitivity to a phase discrepancy between Left and Right. I find that with a 25fps progessive 3D source, introducing a timing discrepancy in the displaying of Left and Right of half a frame, or 20mS, leads to a distracting watery mirage-like effect for my vision. (This was without shutter glasses. in fact I used cross-eyed viewing on a 2D 50p display.)

I have seen the watery effect with commercially released 24p 3D Blu-rays displayed on my Panasonic 50" plasma TV with 120Hz shutter glasses.

These effects involved in-phase capture displayed out of phase. Presumably, out of phase capture with the same time lag, could look worse depending on the movement involved.

I have even seen a watery effect, occasionally, at public cinemas. At the cinema with 144Hz RealD, the vision for the two eyes is slightly out of synch. Each of the 24 frames per second has a Left version and a Right version. Each version flashes on the screen three times to reduce flicker (and to minimise the timing offset between Left and Right). The pattern would be as follows: L1R1L1R1L1R1 L2R2L2R2L2R2 L3R3L3R3L3R3

The left eye is exposed to the new frame slightly before the right eye. Prima facie the display discrepancy is 1/144th sec, or 6.94mS. That is merely a lag in displaying the stereoscopic visual information that would have been captured [or simulated] with correct phase.

I'm planning a little experiment of simultaneously capturing at 50i with an old Canon HV-20 camcorder and at 60p with a Sony TD10, to capture some hyperstereo video. That will result in a varying phase lag at 10Hz when using the separate camera streams for 3D in the Vegas Pro time line. If rendered at 24p, I wonder how noticeable the variation in the capture phase will be. If the scene is only slow moving I'm hoping I might get away with this arrangement, despite the gross difference in frame rates! Of course there will be differences in camera resolution and colour.

MLXXX-

Simply put, you can't mix frames of video that are not quantized to the frame. But shooting with two cameras, each with it's own clock, the videos may have frames that begin at different points in real time. Therefore the two cameras have to be genlocked so that both cameras must have each frame start at the precise same moment in real time. If you don't you cannot edit the video properly. But with digital editing you can slide a video clip and match up the frames in non-real-time in post on the timeline. So actual camera sync in the field becomes unnecessary. In non-3D multi camera joining video with non matched frames will result in a glitch in the cut or a flash on the screen and then the picture will roll until the monitor clocks can sync up to the new clip. Some monitors are faster than others to correct. In very fast responding monitors the edit will look strange and is often referred to as a "Hard edit". In 3D the same is true but the result in the pairing is, as you stated, a watery effect with the edges of the objects in the scene becoming a blur. Since the pixels are now mismatched in time, but the paired clip has a proper sync. I suspect the edits will happen without a glitch but I really don't know because I have never tried to edit with mismatched frame quantization for 3D.

As a degreed and licensed broadcast engineer, I can tell you with certainty that any video that is edited with mismatched frames in the cut points, will not broadcast and would get rejected by any TV station here in the states. As a engineering consultant, I have had my share of jobs where I had to repair videos that were edited with video that was not frame synced in the edits and caused little glitches here and there. You can create all the interesting special visual effects and may get it to play on a very forgiving TV set, but if the technical aspects of the sync are not legal, you may never see it played anywhere else. Bottom line- Don't mess with the engineering of the video synchronization to create a visual effect if you expect the video to work. Stick to basic pixel manipulation for all your visual effects and leave the synchronization alone.

As for doing stereo video with two mismatched cameras- I have tried it and with lots of work, really lots of work, you can get it to work. The big effort will be in getting both frames to match. Forget about zoom as the zoom rates will be different but you may get it to work with static lock down shots. Vegas Pro 3D stereoscopic adjust auto correct will be your friend here. But you may see significant frame cropping. If the lenses are not the same focal length with respect to the imager size (The CMOS or CCD size), you will also need to work hard at getting the 3D to not cause eye strain. Good luck!

Today, video recorded in the pure digital environment ( no tape) has obsoleted the need for the Time Base Stabilizer, the Time Base Corrector for tape recorders. But still necessary are the Frame store synchronizers used by TV stations so that video from various independent sources can be frame synced for broadcast and before that, for basic recording, editing and playback on a TV.

Don,

I agree that you have to adjust the video in terms of frames - and yes: it is important to use quantitize to frames as on (otherwise you may end up with crazy effects, for example if you render the footage the rendering may result in footage that is not readable by the player any more).

And yes, you have to adjust the two clips before you pair them on the timeline - or at least you have to check if such a movement for some frames is necessary or not. I have here wonderfull examples why it is necessary - a steady motive like a house, and then a bird flys through the picture. You see that there is an adjustment required, for some frames. Or when the wind blows in a tree - and the tree moves different since there frames are not adjusted.

For sure that has to be done. It is similar to a multicam editing. I do it exactly that way as you find it in the help fill - put the tracks above and make the higher track transparent, and then adjust the frames. That is some of the reasons why the 3D postpro with paired camcorders is significant more work compared with our compact cameras (since here that does not happen).

BUT are we not talking about another point? If there is a significant drift in synchronisation, does that not mean that we end up with different starting frames all the time? And if the camceras drift and go out of sync - must that not mean that the length of the footage is not similar? Better said - the time distance where the frames are taken? Because if there is a shift in the length of a frame, then you are not able to adjust that in Vegas. That should work only for a similar length of the frames, if they are shifted +/- x frames in the timeline.

The I am not sure about that - since we cannot disabel quantitize to frames there does the shift not result in footage that we cannot align anymore?

I had another look to Mendiburu, because that confuses me (for me it is granted knowledge that we have to avoid the missynchronzation).

And here I became aware that there is a unit called 3D Lanc commander by Michael Starks, what seems to be able to actively synchronize cameras using Lanc.

There is also a nice publication (I had some trouble to downlaod), but you find following comments:


http://www-e2.ijs.si/3dlancmaster/Fi...ster_paper.pdf


To me that would mean that I need synchronized camcorders, otherwise the disparity in frames (!!) become too high. And a correction for frames with different length, or if they start at different points in time, would only be possible if we stretch/compress footage in the timeline - what is a tough exercise (but could be done if the behaviour is linear).

But I agree that this effect is much smaller, compared with what Don described above. But I think it is here in addition!

That would be a preview issue and maybe not the worst part. But what happened to me recently was following: I did a 3-camera MVC editing (2x TD10, 1x Z10K, all in 1080 50i). To adjust the videofootage without quantize to frames - and ended up with jumps in the timeline, in terms of the start-point of a frame. Then I enoded that with the Sony MVC-encoder - and my 3D Prodigy was not able to continue playback that playing 40 minutes, and stopped some time.

I think the encoder was not able to handle that in a correct way, and the player failed to to playback that correct. So the real danger is that the production will fail in the final product. So to my opinion, we should stick to enabled quantize to frame to get full frames.

Fine - it was new for me too. I would like to know a source where we can order that unit... maybe here? It is listed at least for 450 US$.

http://www.3dtv.jp/

I have emailed them.

Happens also here in good old Europe!

That should never happen if following the procedure correctly. Even at high shutter speeds, the exposure of, say frame# 75 on each matched clip properly lined up quantized to frames will show a sliver of the frame, the rest black during the exposure. Both clips should show the same sliver of the frame but maybe at different locations in the frame #75. If you could sync the high shutter speed then the match would work.

Frank, I have not much experience shooting genlocked cameras at non-standard shutter speeds, like 1/1000 sec. While still photographers often select different shutter speeds to stop action, videographers rarely worry about this and stick to the standard 1/60. I fall in this category. Your requirements are not normal and as such may require special needs in editing such as a sync of shutter speeds. This has nothing to do with frame synchronization.

I do agree with you that I don't think the LANC Sony Protocol can sync lock shutter speed on a video camcorder. I believe it can on some still cameras, if I can trust the amateur blogs I have read as truth.

Wolfgang- If all you want to do is shoot wide stereobase video with the TD10's all you need to do is follow what I have successfully done for the past 9 months. Much of what you are doing, I believe, is exploring directions that don't make much sense in meeting minimum needs and is a waste of money such as buying many camcorders in the hopes that two will sync together. As I said, you can sync up both TD10 camcorders using the methods for pairing as outlined in the Sony Help section and when you follow this, you don't even need a Lanc Shepherd or other Lanc controller. All the controller does is make the operation easier and permit a pretty close zoom function that can be tweaked in the 3D stereo auto corrector. I shot quite a bit of 3D wide stereo base before even owning the Lanc Shepherd. Many stereographers have also done lots of work without the need to genlock the camcorders with Lanc controller.

With the Lanc Shepherd, I have learned that it is good for these functions-
1. turning on both camcorders together.
2. start recording both camcorders, close to the same time ( usually within 2-3 frames )
(1 & 2 are usually performed at fairly close to the same time)
3. Zoom is performed in very close sync as long as both camcorders are turned on with the zoom at full wide or full telephoto, against the stop, from the off position.

I was disappointed that the Lanc Shepherd does NOT perform synchronization of:
Exposure, white balance, focus, shutter speed. I don't think any of these is supported by Lanc for camcorders as it is not a supported LANC function per Sony. Also, Lanc does not truly genlock the slave camcorder's clock to the master. I know this for a fact as Sony's Lanc protocol only supports start stop of both recorders. Sony Lanc does support VITC ( vertical interval time code ) but the consumer camcorders do not. The Sony HVR Z1U LANC does support VITC and I designed and built a LANC time code converter to make the LANC time code compatible with LTC ( longitudinal time code) so that I could use the Z1U clock as the master time code generator for my broadcast cameras to lock onto in a genlock required shoot. I believe that the professional version of the TD10 DOES support LANC time code but we are not using these so it is not important. Besides, we can use the clap board method to sync the two as I described earlier. Time code is not required for 3D pairing.

OK, I now understand, Frank. What you are doing is indeed unique. As such your raw clips have to be spot on in the raw state. My procedure is far more forgiving in that I can do tweaking and triming in post for the edit. For paired clips I do the following for a project.- I build a VEG file of just the clips on left and right timelines, left on top with audio. I next line up the audio sync point and then trim off the excess at each end so both are the same length. Next, I pair them and then apply a keyframe set for all the auto correction, usually this means a set in the beginning, holds until the first frame of the beginning of a zoom, again at the last frame of the zoom and then holds until the next zoom. I don't worry about horizontal disparity adjustment until I slice out the part of the paired clip that I need for my edited program. Then I make the horizontal disparity adjustment in order to make the edits flow comfortably. I see that you do none of this as you are working with the original raw clips indefinitely. Therefore you do need to make your adjustments at the camera level since we don't have a 3D live switcher / pairing with hardware correction. Yes, these are made for live 3D events like ESPN. Very expensive! I've seen your laptop setup and it is impressive. Plus you have had to jump through some hoops to get there. But, most of us are not trying to attain that goal and are looking to generate a recorded production.

BTW- the Vegas pairing process does not destroy the original left right clips. It works with a sub clip as the paired version. Some people may not know this and worry otherwise.

The process I described above is what I do to get ready to edit the story. If I were still in the business, it is something I could shove off to my intern assistant editor as it is routine, mechanical, and requires no story telling skills.

Not long ago, while doing another project, I ran into a bug in Vegas that prevented me from proceding with the preliminary step. I could not copy and paste the paired clip from one instance of Vegas to another for the story. When I tried, the pasted clip lost its pairing and was a real mess. Wolfgang found the work around and reported it to Sony as a Bug. When I was at NAB I had access to the Vegas software engineers and reviewed the bug to them and they are quite surprised they hadn't seen that before. I was assured that it would be corrected in the next release but I'm not sure it made it to Build 683. I had to remove build 683 because it crashed so much. I may try to reinstall 683 again now that I'm using a new C drive. Right now my b595 is 100% stable and never crashes.

Don, that is all now anwer to the fact that the two camcorder run at different frequency of camcorder's internal oscillator. That means that the start point of a frame will not stay constant for both cameras. And if the startpoint of the frame has shifted away, I think you cannot correct that in Vegas or another NLE anymore.

The oscillators are indeed running at the same frequency and do drift over time but not significant enough to be detected unless you are recording for many hours. The longest event we ever genlocked was over 9 hours and all 12 cameras at the event remained accurate meaning each camera's clock read the same time at the end of the project. However, as I said, LANC is not genlock but the internal clocks are crystal controlled and as such have a drift that is too small to be of any significance for these requirements. We're talking 9 decimal places of frequency accuracy here. The frame rate constant is considered close enough if we maintain 3 decimals of accuracy. 23.976
What is different since we cannot genlock is the point where the clip starts. As I stated, I see a variance of 2-3 frames in my projects from one shoot to the next not for drift, but rather the point of start record with the trigger of the Lanc Shepherd. Once both cameras are recording, adding the marker for reference is then indicated. If you don't have the marker, your alignment of left and right clips will be from the beginning of the clip. With the marker, you ignore the beginning and line up the markers and now you have your sync point. Slice off the video before the marker. Go to the end of the clip and see that the ends are not aligned, maybe are but trim off what is excess on the longer one. Now both clips are frame matched and everything in the frame is matched excactly, regardless of whether your clip is 5 seconds long or 5 hours long. The number of frames in each will be identical.
Again, the Lanc control unit is not a genlock. It is a simple controller but is not frequency accurate. It is just a switch. The frequency accuracy remains in the cameras. All genlock ever does is set slave cameras to a single camera master, so the clocks are matched to nine decimal places. This is why the start point by the LANC controller may drift 2-3 frames but this is just the start point, not that the fps will drift around. Once started recording the recording progresses at precisely 23.976 fps throughout the clip.

BTW- I am now in process of pairing 112 video clips of various lengths for my Valley of fire project. Every one of them uses the pairing methods described by Sony. They vary in start point by 0-6 frames with the average difference of 2 frames. After aligning to the marker, every one is frame accurate to the end where I also have a second marker. I recall in the beginning when I first got my Lanc Shepherd, I tried to shoot with no markers and ran in to the clip length difference. It was then, that I realized the Lanc controllers do not genlock They are just a switch.

Yes, I share your point that LANC is not a genlock. With lanc you would not be able to run something in sync for 9 hours - cannot imagine that.

And also yes, the controller start the camcorders, and I see also a difference here in some frames in the startpoint - we discussed that this must be corrected.

Maybe you are lucky and have a pair of TD10s that are wonderfull in sync too. But if a shift takes place, you can calculate and measure how long it takes you that you come out of sync. Have you measured how long it takes your pair to go out of sync? I have done so and know that my numbers are not great. 10 minutes or so until I have the 1 ms.

Since you have to stick to full frames you cannot correct that shift by adjusting that by another half frame, or 0.33 frames or whatever. But that is what would be necessary after some shifting. The best what you can do is to reset the camera and start again. Or you have a pair that gives you more then my 10 minutes.

I'd say that the record switch latency of the Sony is larger than the Canon, BUT- I'd question how you were able to determine the canon started at a half frame point. Was that an exaggeration to make a point or do you have empirical data to back that up? This is a technical impossibility because no camcorder manufacturer would ever design a camera that did not use a vertical interval record start point. I'm not even sure how one could design such a circuit. My guess is your method of measurement is in question, or, did you just assume that? Hint: you could verify this with a waveform monitor set to 2 Field display. I think Tektronics has a trigger hold to do a snap shot of the first signal start point if you have access to such high end equipment. I have several WFM's here but none have the trigger hold to be able to see that. Used one in school a few decades back.



No doubt. They all drift over long time. Never questioned that. We even had a camera to shoot in a ball park in the sun all day was different than the ones in the shade but the drift was only a couple frames after an all day shoot with genlock once in the early morning, and inside an air conditioned building. After that experience I chose to re-genlock the cameras at lunch break. Temperature differences is a primary cause of crystal oscillator drift, but as I stated before these oscillators are generally much more accurate than is necessary to maintain long periods of frame accuracy. We're not talking about new science here. This stuff has been industry standard for as long as I've been associated with broadcast engineering and I got my degree in 1969. In high end TV stations, the house sync generator is in a temperature controlled oven but in field shooting, not necessary since genlocked cameras never seem to be needed for longer than a day.

Now, if you can show me a camera that shoots one time a 100 second clip that has 2997 frames in it and the next time it has 2875 frames and the next time it has 3005 frames, then I will say you have a camcorder that is in dire need of repair.

I thought we were discussing science. As friends, I trust you believe in what you see.

Meanwhile- I need to get back to pairing and keyframing my clips. But, please consider this- I don't have the problem. When I do, I'll let you know.