Rolling Shutter CMOS Artifacts

It's one good component of a the test.

I would also pan by a series of verticles like a picket fence to check as well.

That way we can see the hard edges of the fence as well s the background in between the fence.
 
ESTEBEVERDE,

When you nodally pan a camera, everything in the frame will motion blur equally, both the "pickets" and the stuff between them. Perhaps you mean dollying past a fence? This would apply more motion blur to the foreground objects than the background material. But then I'm not sure what that would prove vis a vis the assertion that there is resolution loss while panning.
 
No... just the fence.

The foreground will be moving across the frame at a slower relative speed and because it is a picket fence it would be easier to see how much blur we are getting.

If this is all blur the picket fence should blur less than the back ground.

On a dolly move the relative speed would be the same as it is linear and not an arch...
 
ESTEBEVERDE,

Motion blur is a 2D phenomenon. As measured in pixels, a nodal pan will create equal length streaks (at a given position in frame because of lens distortion) on everything in frame. With a lateral dolly move (no camera roation), foregound objects will streak a certain amount while obects approaching infinity will appear to stand still - hence no motion blur.
 
Hmmm....

O.K.

Let's look at it like this.


Do a "reverse pan" where the subject moves and not the camera.

Put the camera in the center of a gigantic cd.

Now, step three feet away.

Start at the right side of the frame and run just to the left.

Measure your distance and speed.

Now... step 200 feet away and do the same thing.

Start at the right side of the frame and run just to the left.

Measure your distance and speed.

You have to move much faster and farther the further you are away from the camera.


And.. on second thought... it's actually the same with the dolly move.

Foreground objects have slower relative speed but cause more blur....
 
matthew77 said:
ESTEBEVERDE,

Motion blur is a 2D phenomenon. As measured in pixels, a nodal pan will create equal length streaks (at a given position in frame because of lens distortion) on everything in frame. With a lateral dolly move (no camera roation), foregound objects will streak a certain amount while obects approaching infinity will appear to stand still - hence no motion blur.
Click to expand...



Another way to look at it is will an airplane traveling at 500 miles per hour 2 miles away cause more motion blur or one 2 feet away?
 
What about slow-moving shots?

Think of movies such as Elephant, or Magnolia.

Some of those shots are at the pace of someone normally walking, for a rather soothing effect. Some of the shots are level, then "dip" in to dutch angles, then come back.

Would a rolling shutter have a serious effect on these types of shots?

(Sorry to temproarily hijack the thread, BTW)
 
The rolling shutter becomes most noticeable if you reverse direction. Gliding smoothly will not reveal much in the way of artifacts, if anything. A quick reversal will make the artifacts very obvious. A handheld shot on an HV20, for example, is much jigglier than the same handheld shot on a DVX. As the handheld motion results in frequent reversals (camera is not gliding, it's more "bouncing") then the effects of the rolling shutter become more obvious.

If your camera movements are highly controlled and smooth, the rolling shutter artifacts will be masked. If the camera is moving in a more uncontrolled way, "Cops"-style, the rolling shutter will be much more problematic. The ultimate nightmare scenario for a rolling shutter is a high-vibration scene.
 
I've added tilt test results showing expansion and compression of the image to that first post in this thread...

http://www.reduser.net/forum/showpost.php?p=134798&postcount=1

It would be a real asset if anyone with an EX1 could do similiar sort of work...

For instance, it's deductable from my first pan test that an image subject in motion moves from one screen edge to the other in approx. 18 frames will skew 3% of the 4K width. Charting out an EX1 would give us more information about skew in general and help compare the RED to EX1 on equal terms.

Regards,

Jim Arthurs
 
Wow, that's an eye opener more than any other artifact. So have you determined the minimum speed an object has to pass between to points to cause noticeable artifacts, tilt, or pan?

Also, Jim, can you send me details on how I can set up this experiment with the EX1, or if you're ever down in Phoenix AZ area?
 
Last edited:
Hi Steve... all you need is a sheet of black foam core board from a Michaels or Hobby Lobby... then some white chart pack tape 1/16" inch or rip off a thin strip of white grip tape.

I wouldn't use the dots I used on the RED test, upon reflection, I'd just do a series of long thin strips an inch apart running up and down the whole length of the board. Then shoot it both horizontal for pans and vertical for tilts. Do one pan very slow with the tilt locked off to be a base reference. Then do another where you whip through frame each direction. I'd let each pan motion go completely off the target in each direction and try and keep the speed constant through the pan. Do different speeds.

Then repeat the same for tilts; a base tilt recorded first, very slow, and then your faster ones.

If you wish to shoot something like this, I'd be glad to do the post on it for you so we can share the results... let me know PM and I'll set you up some FTP space to upload the files.

Oh, and you can kind of figure out how much skew at different speeds... if 3% width with a pan at 18 frames from left to right, then 1% skew with a pan that's 54 frames... so a 3 second or longer pan time for an object crossing the frame (at least for the RED) shouldn't have very much skew at all.
 
Yes, full recorded 4096 wide. 2K would be half as bad, maybe less if I understand that the reset is not directly proportional... don't quote me on that.

BTW, here's an EX1 test that has just come to light by Eric Pascarelli.

http://dvinfo.net/conf/showthread.php?t=112362

While there's some slop in my test, his is very cut and dried with the outcome when you measure the frames in Photoshop... 2.2% width of frame skew in 17.6 frames for a subject to cross the frame.

My RED test was approximately 3% skew in 18 frames. My pan speed isn't perfectly constant like Erics. It's clear even with some error that these two cameras are very very close. It's possible that the EX1 has a little less skew than the current firmware build on RED.
 
Thanks Jim, When I first saw your thread and results from your tests, that was my immediate thought. It sure looks just like the EX1.
My thought is there really is not much they can do about this issue via software.


If there was a way, they surely would of implemented it from the start.

I did hear about RED's upcoming 10% improvement.

So, based on those results from Eric, It appears when shooting at 24 FPS, a 3/4 second constant pan speed measure from edge-to-edge will yield 2.2% frame skew.
So, I imagine if you kept this speed to at least 2 seconds or more, there may be little to no visable skew. I'd like to see this also done for the tilt axis.
 
Last edited:
Thanks for the reference to Eric's test.

Aside from the skew, the pictures at high shutter speeds look pretty sharp. Barry Green mentioned that the picture lost sharpness during pans, but maybe it was just due to motion blur?
 
I'd have to get more time on it to find out. There are several posts raising questions about what happens in motion; Noel Evans just posted something. I am not sure what was going on, but something seemed "off". There was definitely some motion-induced res loss happening, and it *seemed* to be moreso than would be attributed to motion blur. But I'd have to get some more time on an EX1 to get to the bottom of it.
 
You must log in or register to reply here.
Back
Top