Sample Clip Below:

I've spent a couple of months and too much money already trying to figure this out myself. So I hope someone can tell me if what I am trying to do is even possible.

I assumed I would always be able to get crisp clear video with a HD camcorder. I am trying to take video of garden scenery which consists primarily of many slow pans of 180 degrees or more.

However, as I found out too late, this is not easy. There is a lot written, and a lot of explanations about panning and motion blur, especially with the new consumer AVCHD camcorders. Everyone seems to have a different story.

I am taking a slow 180 degree pan on a tripod, with OIS off, with my HF-10 at 1/500 (1080-60i)on a bright day, and it shows no detail on the plants at all. I log and transfer into FCE and export as a QuickTime Movie. When the camera is still, the quality is excellent. When panning slowly, the detail disappears entirely. This sample is using autofocus, but infinity settings didn't help. In fact, even the "Easy" setting was pretty close to what this is. The faster shutter speed didn't help but very little. I've tried all the other tips I've read on the web to no avail.

The only thing I don't really understand are the de-interlacing issues. I tried 30p, and just about every other setting that exists.

Here is a Hi-quality 55 MB Fast Start QT Clip. It should auto start about half way loaded - 30-40 secs on a cable modem. Please look and tell me is this the best I can do. It sucks as far as I'm concerned.

SAMPLE 1/500 CLIP (http://www.palmpedia.net/wiki/index.php/Slow_Pan_Test_1/500)

Is it the camera, the AVCHD format, the software (FCE), my MBP, or is this not something HD video can do? If I spent a few thousand dollars on a better camera, would I still be unable to get any detail in a slow pan?

What would a pro do? Is that something I could do for under $10,000. I already have a decent Mac and FCE. I would hate to spend $5,000 on a camera and still have blurry pans.

I think your intention of testing panning shot pretty much says it all. encoding bit rate looks high enough to recover motion estimation residue. however the intrinsic interlace artifacts are there because of 1080/60i. De-interlacer is working hard but unfortunately there is no way of eliminating the artifact, plus the de-interlacer also introduces aliasing.
Have you tried progressive mode?

Deinterlace using "bob" method, should be much better. Or shoot in 30p.

your post in the other thread says compression bit rate 2000kbps. maybe On2 VP6-S is for web posting, 2Mbps for 1080/60i video is a pretty low bit rate. details are being quantized away in this case.

That particular test clip is done at a high bit rate to illustrate for you guys the blur, and not confuse it with compression issues. It's around 9,000 kpbs and represents pretty well the extent of the problem. I would have thought 1/500 would have taken away some of the blur.

I have tried 30p, but it doesn't seem to be any better. In fact, it looks a little worse.

So is this blurring a result of the 60i format? I've tried several de-interlacing programs. Would a 60p camera solve this problem? Or is this an inherent problem with AVCHD? Or is it CMOS that is doing it? Or just the result of using a consumer camera processor and glass.

In other words, what do I need to do to solve this problem? Do I need a better camera? Can high end cameras produce the quality I need to do video like this of gardens? Is this issue solvable?

I've attached a screen shot of stop action from the original uncompressed footage, in case that helps you guys identify anything.

Have you tried looking at the original mts video file (using the camcorder directly hooked up to a HDTV using HDMI)? Your clip seems to be a re-encoded / converted file, so it may be hard to know for sure if that may play a role at all.

your post in the other thread says compression bit rate 2000kbps. maybe On2 VP6-S is for web posting, 2Mbps for 1080/60i video is a pretty low bit rate. details are being quantized away in this case.

FYI - My intent is to eventually place 540p (at 2000-3000kbps) or 720p at little higher bit rates on my website. Both .mov and .flv compression techniques are up to the task these days. I have seen some incredibly clear underwater 540p clips at 3000 kpbs that just blew my mind.

But I would have to start with some very clean original material.

Have you tried looking at the original mts video file (using the camcorder directly hooked up to a HDTV using HDMI)? Your clip seems to be a re-encoded / converted file, so it may be hard to know for sure if that may play a role at all.

Kalak,

Thanks for bringing up that point. And thanks to everyone who is taking the time to view the test clip and comment.

That test clip is almost identical in the "blur factor" to the unadulterated video on my computer before any compression. The blur is there before I compressed it.

However, when I play it on my 42" 1080p LCD Sony directly out of the camera there appears to be less blur. I don't have a mini HDMI, but I used component. I intended to mention that fact if this discussion got going, but I have read so much about motion blur problems with the AVCHD format and the HF-10 & 100 that I wanted to see if you guys agreed that the camera is the likely culprit.

But let's assume for the sake of this discussion (and the following point, that the original .mts is of decent quality. Maybe I am doing something stupid. I Log and Transfer the .mts file into FCE. I then "add it to the queue." And from there I export as a QuickTime movie with the only option (Make movie Self-Contained) checked. I have tried using Voltaic as well. Same results.

If I play these exported AICs with QT on my 17" Mac Book Pro, I get the blur. Is this file still interlaced? From there I have tried everything. I've sent the AIC into Sorenson, On2, Episode, all with the same results. But since I am feeding in blur, it's no surprise. But exporting and compressing directly from FCE yields pretty much the same results.

Am I missing an important part of the work flow?

Hi Dypsisdean,

It there any way you can post the original raw .mts file from the camera, so we can have a look at it?

Here's my layman's take on the panning softness problem. I have enough knowledge to be dangerous, but not enough to guarantee that everything is precisely accurate.

AVCHD (and other MPEG based compressions schemes) use a sequence of frames (I,B and P) that have different amounts of compression applied. This sequence, called a GoP, or Group of Pictures is quite long, I believe it is 16 frames. So the first frame in the GoP is an I frame, or Initial Frame, and is the least compressed and has the most detail. Much of the information contained in this I Frame is used to help rebuild the pictures from the subsequent, lower detail frames in the GoP. The B frames, or Bi-Directional Frames have more compression applied than the I frames do. They basically consist of information that was not found in the original I Frame and they allow playback in either forward or reverse. The P Frames, or Predictive Frames are intended to provide motion information, used in reconstructing the frames. Again, the P Frames are based on detail that changes from frame to frame, but there are limits to how much detail can be stored in a P Frame.

When panning, very little information remains identical from one frame to another. The only way to preserve all the detail would be to shoot all I Frames, but this is not practical at these data rates. Therefor when recording with AVCHD, only one frame in 16 (if this is the correct GoP length) has the full detail of the image.

On playback, all the I, B and P frames get reconstructed as quasi I Frames, from the limited detail that is available. So in panning shots, you see much more blurring than you would if all the frames were I Frames.

In broadcast camcorders, most formats shoot all I Frames, so there is no problem in panning and so that editing is much less demanding on the computer being used.

rbouch,

If your "lack" of knowledge is dangerous, then my lack of knowledge is "lethal." :) If your analysis is correct, what does that mean? Since the clip plays fairly well out of the camera, do you think two things going on?

1) AVCHD shortcomings that don't show up as much before conversion.
2) Software that's unable to convert AVCHD adequately.

Willc,

Hopefully I've placed the original .mts in a place that is downloadable for anyone that is still with me on this.

Go back to the top of this page this page. I hope this works for you.
SAMPLE 1/500 CLIP (http://www.palmpedia.net/wiki/index.php/Slow_Pan_Test_1/500)

My biggest question, and one that would help the discussion, would be the following:

Can anyone present a clip from any AVCHD camera of a similar pan without blurring???

(I put this in your old thread too)

All HD (or HDV cams) on some sort of mpeg compression will give you a bit of blur on panning, but the avchd cams appear to be worse. There are a few things you can do to lessen the blur... pan slower, don't use zoom or OIS, but generally speaking the MPEG2 cams (HDV) present less blur on pans.... so if you're that disappointed, then you may want to look at HDV.

(I put this in your old thread too)

All HD (or HDV cams) on some sort of mpeg compression will give you a bit of blur on panning, but the avchd cams appear to be worse. There are a few things you can do to lessen the blur... pan slower, don't use zoom or OIS, but generally speaking the MPEG2 cams (HDV) present less blur on pans.... so if you're that disappointed, then you may want to look at HDV.

Bigbarney,

Does that mean if I purchased a camera that didn't use mpeg compression (are there any???) would that solve my problem? If I used sophisticated H.264 compression after the fact, would that bring the blur back?

In reference to the explanation of GoP by "rbouch" I found this to be very interesting, and a solution to the apparent limitations of AVCHD when it comes to motion and panning blur. I think, in my limited knowledge, that it confirms what "rbouch" was explaining.

I haven't checked at what price it comes at, but I'm sure it is expensive. Does anyone care to comment?

AVC-Intra (ftp://ftp.panasonic.com/pub/Panasonic/Drivers/PBTS/papers/AVCIntra%20FAQs.pdf)

Does anyone care to comment?


I wouldn't give it too much weight.... right now anyway. It appears to be a Panasonic document on a Panasonic idea. It sounds great but I would want to see it from an independent before I took it seriously.

So far.... avchd has not gone anywhere in the semi pro arena. Sony just came out with a new line of semi pro cams.... and they're ALL hdv.... NONE of them are avchd. To this point avchd is little more than a consumer level codec/format. Whether that changes is anybodies guess... I'm betting it will ..... SOME DAY, but no time soon!

Bigbarney,

Does that mean if I purchased a camera that didn't use mpeg compression (are there any???) would that solve my problem? If I used sophisticated H.264 compression after the fact, would that bring the blur back?

Using any codec/format AFTER the fact won't help, and there are no hi def cams (worth anything anyway) that don't use an mpeg based compression. I know there was one cam out about a year ago that worked on the mjpeg principle.... but the reviews were not very favorable.

Don't forget though... the cam, lenses and other such items have an effect too. If you get a semi pro cam like the A1 and learn how to use it.....

And HDV may be better in terms of panning, but it has its draw-backs as well so you need to do your studies and pick the least of the evils as you see them.

When panning, very little information remains identical from one frame to another. The only way to preserve all the detail would be to shoot all I Frames, but this is not practical at these data rates. Therefor when recording with AVCHD, only one frame in 16 (if this is the correct GoP length) has the full detail of the image.

That's not quite true, otherwise even a little bit of shaking would cause huge data loss in non-I frames. All MPEG compression schemes involve some sort of motion estimation (http://en.wikipedia.org/wiki/Motion_estimation) to figure out the motion vector between two frames. So, as you pan slowly to the right, most of every frame is extracted from the last I frame, and only the new pixels on the edge have to be encoded... and that's why you still get good compression in panning shots.

Now, this is still obviously worse than if you had a static shot, but a lot of information still carries from one frame to another.

OK. did a few tests and I tend to believe it's more display-dependent.
First thing I checked was the bit rate. this video clip was shot as 1080/60i @ ~16Mbps.
For a h.264 encoder with full HD resolution, 16Mpbs seems adequate to preserve good amount of details in the image. to verify my point, I view the clip frame by frame. shots did look fine, lots of details preserved. especially when no panning, frame-by-frame looks almost identical to normal speed.
However, between pauses, normal-speed playback looks blurry. I start to think toward the display side. At this moment I am using PS3 as decoder and de-interlacer (set display mode to 1080p), the TV is Sony 46" Bravia XBR2 which is LCD display with 6ms response time. Slow response time is also a possible cause for motion blur.
Decided to to a quick comparison, I connect the ps3 to my Sony 70" KDS-R70XBR2, which is SXRD with 2.5ms response time. The outcome is astonishing yet as expected---motion blur is almost gone. I can only say the difference is fairly noticeable if not day-and-night.
Unfortunately I am no fan of plasma display, otherwise we could've had more tests to verify motion blur in the particular case is mostly caused by display response time.

I guess you view most of your clips on your MBP, but what is the native resolution and the display response time? I suggest you find a tv with quick response time (<4ms) and see for yourself.



BTW I also tested de-interlacer by ps3 and the tv. Apparently, ps3 is doing a much better job.

That's not quite true, otherwise even a little bit of shaking would cause huge data loss in non-I frames. All MPEG compression schemes involve some sort of motion estimation (http://en.wikipedia.org/wiki/Motion_estimation) to figure out the motion vector between two frames. So, as you pan slowly to the right, most of every frame is extracted from the last I frame, and only the new pixels on the edge have to be encoded... and that's why you still get good compression in panning shots.

Now, this is still obviously worse than if you had a static shot, but a lot of information still carries from one frame to another.
As I said, the P Frames are "Predictive" Frames. They provide the "motion estimation" information you describe. However, when panning quickly, more pixels change than can be estimated and a reduction in total resolution is the result.

To me, this looks like it's mostly a deinterlacing issue. The problem is that we are all using different hardware, software, TVs etc, and it's really difficult to confirm that we are all seeing and talking about the same thing.

In order to level the playing field slightly, I have uploaded some progressive versions of your video created from the original mts file using Vegas. This should eliminate the deinterlacing from the equation for the time being.

To me these all look fine, with super-smooth motion at 60fps and no blurring. Could you try downloading these and confirm you see the same, and that they are the sort of quality you are looking for?

Ignore the videos as displayed by Vimeo (as they are converted to low bitrate, 30fps), and click the download links on each page to download the original mov files. You'll need to sign up to Vimeo if you're not already a member to be able to do this (it's pretty painless).

Password for each page is 'avsforum'.

720p60 16mbps (need meaty PC/MAC to view smoothly)
http://www.vimeo.com/1697239

720p60 4mbps (less meaty PC/MAC to view)
http://www.vimeo.com/1697270

720p30 4mbps (30fps for comparison)
http://www.vimeo.com/1697288

480p60 4mbps (for lower spec display)
http://www.vimeo.com/1697303

As I said, the P Frames are "Predictive" Frames. They provide the "motion estimation" information you describe. However, when panning quickly, more pixels change than can be estimated and a reduction in total resolution is the result.

this is probably too much into how encoder works, but:
1. bit rate determines how much details to be encoded into I/P/B frames.
2. panning is usually where encoder gets high compression ratio because MB-based motion estimation yields low residue.
3. when panning quickly, resolution doesn't matter because captured images are blurry anyway.
4. when lost of resolution exceeds preset threshold, in common practice there will be I-frame insertion.

hope this helps.

guess I overlooked it.



However, when I play it on my 42" 1080p LCD Sony directly out of the camera there appears to be less blur. I don't have a mini HDMI, but I used component. I intended to mention that fact if this discussion got going, but I have read so much about motion blur problems with the AVCHD format and the HF-10 & 100 that I wanted to see if you guys agreed that the camera is the likely culprit.



Your Sony LCD has 4-6ms response time, while your MBP probably has 16ms (http://forum.notebookreview.com/showthread.php?t=131574).
think your MBP LCD is the culprit.
also the component cable is no good for high quality video. use HDMI.

when I play it on my 42" 1080p LCD Sony directly out of the camera there appears to be less blur.
Because it deinterlaces using "bob". Have you tried deinterlacing using "bob" as I suggested? Basically, convert each field into a frame by doubling its lines, this would give you 60p. Or just drop every top or bottom field, this would give you 30p.

In what way 30p was worse?

Because it deinterlaces using "bob". Have you tried deinterlacing using "bob" as I suggested? Basically, convert each field into a frame by doubling its lines, this would give you 60p. Or just drop every top or bottom field, this would give you 30p.

In what way 30p was worse?

FYI, top-tier TV manufacturers like sony don't use line-doubling to deinterlace. there is usually "Motion estimation and compensation" scheme involved.

FYI, top-tier TV manufacturers like sony don't use line-doubling to deinterlace. there is usually "Motion estimation and compensation" scheme involved.
Prettly much the same thing. You have 60 images per second in interlaced, you need to generate 60 frames per second for the same motion feel. There are some clever tricks, but still they boil down to "bob" and then build on it. In smooth scenes these clever tricks may combine fields, effectively dropping frame rate to 30 fps, but in fast scenes there is no way of doing it while preserving the same fluid motion, unless they generate fake frames like in 120Hz models, which look pretty awful to my taste.

I am taking a slow 180 degree pan on a tripod, with OIS off, with my HF-10 at 1/500 (1080-60i)on a bright day, and it shows no detail on the plants at all. I log and transfer into FCE and export as a QuickTime Movie. When the camera is still, the quality is excellent. When panning slowly, the detail disappears entirely. This sample is using autofocus, but infinity settings didn't help. In fact, even the "Easy" setting was pretty close to what this is. The faster shutter speed didn't help but very little.

you definitely don't want to use autofocus for this kind of a shot, and infinity focus probably isn't the best choice either... you should be able to zoom in fully to the plant areas in the middle of the yard, off to the side, set focus with the autofocus on, then switch to manual focus without changing the focus point.

have you tried shooting the pan at 1/60th, in shutter priority mode? 1080i is good for this kind of a shot, i would not want to do it in progressive mode.

good feedback in this thread on how display devices can affect the picture quality... i looked at your sample clip on an expensive 24" crt monitor(~$3k back in the day), which of course has zero lag, and it didn't look all that bad... nero showtime, adaptive interlacing set on full... the birds flying across the screen in the middle of the pan looked good.

Prettly much the same thing. You have 60 images per second in interlaced, you need to generate 60 frames per second for the same motion feel. There are some clever tricks, but still they boil down to "bob" and then build on it. In smooth scenes these clever tricks may combine fields, effectively dropping frame rate to 30 fps, but in fast scenes there is no way of doing it while preserving the same fluid motion, unless they generate fake frames like in 120Hz models, which look pretty awful to my taste.

they don't boil down to "bob" to do 60i-60p conversion.
say you have 60i source top/bottom fields coming in as t1, b1, t2, b2, t3, b3
simplest way to convert to 60p is weaving current field with previously buffered field:
(t1, b1), (t2, b1), (t2, b2), (t3, b2), (t3, b3), ....
in this way you get deinterlaced 60p.

they don't boil down to "bob" to do 60i-60p conversion.
say you have 60i source top/bottom fields coming in as t1, b1, t2, b2, t3, b3
simplest way to convert to 60p is weaving current field with previously buffered field:
(t1, b1), (t2, b1), (t2, b2), (t3, b2), (t3, b3), ....
in this way you get deinterlaced 60p.
If they were literally weaved, then you would get combing. If some clever motion estimation is used, then this is not the original image. I am not sure that I would want to watch this sort of video. I watched Sony's MotionFlow at 120Hz with motion estimation and did not like it. Maybe it is just me.