"Rolling Shutter" Skew Tests and Grabs at DVinfo.net

Eric Pascarelli · January 14th, 2008, 12:36 PM

I put the EX1 on a motion control head to try to quantify the amount of skew that is created by the "rolling shutter" on the EX1. (can we start calling it a "scanning shutter" instead?).

A chart with vertical bars was placed in front of the camera (which was at full tele to minimize distortion) the Kuper motion control head was programmed to pan, at a linear speed, precisely one bar per frame of video. I was only able to test 30p, and I would like to test other speeds, but my sense is that it would all be proportional.

What I concluded is that the skew is the same for all shutter settings (I tested 1/2000, 1/250 and 1/60) and that the shutter scans at twice the rate of the camera - that is, if the camera is set to 30fps, the shutter scan takes 1/60 second from top to bottom. Higher shutter speeds are achieved by a "slit scan" of the sensor, similar to the curtain shutter of an old film SLR camera.

Here is how I arrived at that: Each bar on the chart is 110 pixels wide. So we can say that the EX1 is panning at 110 pixels per frame. The amount of skew from top to bottom is 55 pixels - half of 110, and hence half of the duration of the frame.

If you are curious about the equipment I used, go to www.general-lift.com for more info. There are pictures of the motion control head etc. They were kind enough to let me use their gear.

Here's what the grabs show:

No skew - static chart
Skew at 1/2000 (with graphic)
Skew at 1/250
Skew at 1/60

Paul Curtis · January 14th, 2008, 01:08 PM

Eric that's some very good clean results, thank you.

If it moves that much horizontally then you know it will do the compress/expand effect vertically as well. I would have thought that would rule out shooting in a car for example.

I quite disappointed with the results, as indications had suggested (along with sonys marketing) that it would be a lot better than this. Seems like quite a slow read out to me too (1/60th second!)

thanks
paul

Leonard Levy · January 14th, 2008, 01:15 PM

Eric,
Thanks for your careful testing and hard work.

Can you post any video so we can see how fast these pans are in real motion? It's great to know that shutter has no affect on the skew , but I want to know when my pans or motion will begin to be affected.

Perhap[s tghat was not part of your testing though.

Noah Yuan-Vogel · January 14th, 2008, 02:06 PM

very interesting. this is a very well done test. it would be more interesting to see a comparison between different formats since the fact that the speed of a cmos sensor's electronic rolling shutter does not affect the the readout rate or resulting skew has been documented. with various formats, however, depending on how the various frame sizes and frame rates are arrived at in terms of sensor clocking, blanking, etc. there could be a significant difference in the skew effect.

Eric Pascarelli · January 14th, 2008, 02:14 PM

Noah - good point. I will have to try 720 - knowing what I know now I could easily to it on my fluid head, analyzing two successive frames, and won't need to use the moco.

My guess is that it will be the same, but it's worth a try.

Leonard - I will post a quicktime of the pan - I'm just sorting out with Chris H. why I can't upload a small mov file.

Noah Yuan-Vogel · January 14th, 2008, 02:39 PM

it seems that it would be difficult to guarantee that the pan speed will be constant each time without motion control. perhaps there is some object that can move across the screen a known speed? without knowing precisely the speed of the pan/movement it would be hard to be sure of the results especially if the differences are not large in relation to the size of the video frame.

Steven Thomas · January 14th, 2008, 02:44 PM

Eric, thanks for this test.
Can you give us an idea if it's close to the RED ONE numbers below for skew based an image moving across the sensor at a consant speed.

Jim Arthurs' tests show with the RED ONE 18 frames will skew 3% of the 4K width.

This means if shooting at 4K 24P, a constant speed pan (good luck ;) ) from one edge to the other takes 750ms, it can cause 3% skew on the 4K images.

Jim also mentioned if the pan speed was reduced to 3 seconds, there's virtually no visable skew.

Steven Thomas · January 14th, 2008, 03:05 PM

Eric, if you still have the test stuff around, can you do the same thing for tilt (up/down) movement. I'd like to see this measure affect too.

Vince Gaffney · January 14th, 2008, 03:57 PM

So Eric,

Is it safe to assume that since you were long on the lens and in a practical way took the arc out of the equation, we should assume the same results for a dolly - or tracking move? If all other factors are relative - frame, focal length, distance from object etc.

vince

Eric Pascarelli · January 14th, 2008, 04:14 PM

The point of going full tele was to take the arc and the rectilinear lens distortion out of the equation as much as possible.

Even with the arc, the measurements were taken near the center of the image, and from the same part of frame, so the arc is not significant.

Skew should be considered a 2D phenomenon that happens on the flat surface of the sensor - is doesn't matter how you arrive at the motion that's projected on the chip - it could be dolling, panning, zooming, subject motion or some combination thereof. But if there's something that should appear vertical moving across the chip horizontally it will be skewed.

Likewise, as I'm sure is obvious, if you are panning or dollying with a moving subject, the subject will not be skewed and the background will.

Eric Pascarelli · January 14th, 2008, 06:00 PM

So I looked at this footage shot by Jim Arthurs on the RED forums and analyzed two successive frames - 170 and 171.

http://www.reduser.net/forum/showpos...98&postcount=1

(It's the first link)

The movie is low res (800x450) but the proportions are correct. The pan speed there is 66 pixels/frame and the skew amount is 22 pixels.

So in this example, the RED shutter scans the sensor at 1/3 the frame interval, which is better than the EX1 which scans at 1/2 the frame interval.

I'm checking with Jim to make sure that this image is downresed from the full chip rather than cropped. Cropping would reduce the amount of skew in for the part of the sensor that's used.

Matthew Roddy · January 14th, 2008, 07:24 PM

I hope everyone will forgive a very novice question (I've never heard of skewing, so this is all educational for me):
Is Skewing an inherent trait to all DV cameras? or CMOS only?
For example, I've not noticed it in my Canon XL-H1, or my Sony Z1u. Which is to say, I've never looked for it and rarely do fast pans and the trait may very well exist.
I look at the examples and am somewhat awed.

Carl Middleton · January 14th, 2008, 07:37 PM

Quote:

Originally Posted by Matthew Roddy

Is Skewing an inherent trait to all DV cameras? or CMOS only?

It's CMOS only - it uses a 'rolling shutter' (or 'scanning shutter' ;D ), sampling the image from top to bottom. This causes distortion if the image has moved between when the top and the bottom have been sampled. CCD cameras such as the Z1 (which I own as well) do not exhibit this behavior as they scan the whole image at the same time.

Hope this helps!

Carl

Matthew Roddy · January 14th, 2008, 08:37 PM

Quote:

Originally Posted by Carl Middleton

It's CMOS only...

Hope this helps!

Carl

It does. Thanks Carl.
I was HOPING it was in all cameras of this nature - then I would say, "Engh, all cameras do it, so I'm OK with purchasing the EX1."
But since this is something inherent to CMOS, I have some weighing of priorities to consider.

Thanks very much.

Mooho Bae · January 14th, 2008, 08:42 PM

Thank you so much Eric.

I'm interested in the 24P case. Is the scan speed still remained at 1/60 sec, or reduced to 1/48 sec. Would you please examine that?

January 14th, 2008, 12:36 PM	#1
Eric Pascarelli Major Player Join Date: Nov 2007 Location: Los Angeles, CA Posts: 498	"Rolling Shutter" Skew Tests and Grabs I put the EX1 on a motion control head to try to quantify the amount of skew that is created by the "rolling shutter" on the EX1. (can we start calling it a "scanning shutter" instead?). A chart with vertical bars was placed in front of the camera (which was at full tele to minimize distortion) the Kuper motion control head was programmed to pan, at a linear speed, precisely one bar per frame of video. I was only able to test 30p, and I would like to test other speeds, but my sense is that it would all be proportional. What I concluded is that the skew is the same for all shutter settings (I tested 1/2000, 1/250 and 1/60) and that the shutter scans at twice the rate of the camera - that is, if the camera is set to 30fps, the shutter scan takes 1/60 second from top to bottom. Higher shutter speeds are achieved by a "slit scan" of the sensor, similar to the curtain shutter of an old film SLR camera. Here is how I arrived at that: Each bar on the chart is 110 pixels wide. So we can say that the EX1 is panning at 110 pixels per frame. The amount of skew from top to bottom is 55 pixels - half of 110, and hence half of the duration of the frame. If you are curious about the equipment I used, go to www.general-lift.com for more info. There are pictures of the motion control head etc. They were kind enough to let me use their gear. Here's what the grabs show: No skew - static chart Skew at 1/2000 (with graphic) Skew at 1/250 Skew at 1/60 Attached Thumbnails

January 14th, 2008, 06:00 PM	#11
Eric Pascarelli Major Player Join Date: Nov 2007 Location: Los Angeles, CA Posts: 498	Red Comparison So I looked at this footage shot by Jim Arthurs on the RED forums and analyzed two successive frames - 170 and 171. http://www.reduser.net/forum/showpos...98&postcount=1 (It's the first link) The movie is low res (800x450) but the proportions are correct. The pan speed there is 66 pixels/frame and the skew amount is 22 pixels. So in this example, the RED shutter scans the sensor at 1/3 the frame interval, which is better than the EX1 which scans at 1/2 the frame interval. I'm checking with Jim to make sure that this image is downresed from the full chip rather than cropped. Cropping would reduce the amount of skew in for the part of the sensor that's used.

January 14th, 2008, 01:08 PM	#2
Paul Curtis Major Player Join Date: Mar 2005 Location: canterbury Posts: 411	Eric that's some very good clean results, thank you. If it moves that much horizontally then you know it will do the compress/expand effect vertically as well. I would have thought that would rule out shooting in a car for example. I quite disappointed with the results, as indications had suggested (along with sonys marketing) that it would be a lot better than this. Seems like quite a slow read out to me too (1/60th second!) thanks paul

January 14th, 2008, 01:15 PM	#3
Leonard Levy Trustee Join Date: Feb 2006 Location: USA Posts: 1,684	Eric, Thanks for your careful testing and hard work. Can you post any video so we can see how fast these pans are in real motion? It's great to know that shutter has no affect on the skew , but I want to know when my pans or motion will begin to be affected. Perhap[s tghat was not part of your testing though.

January 14th, 2008, 02:06 PM	#4
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	very interesting. this is a very well done test. it would be more interesting to see a comparison between different formats since the fact that the speed of a cmos sensor's electronic rolling shutter does not affect the the readout rate or resulting skew has been documented. with various formats, however, depending on how the various frame sizes and frame rates are arrived at in terms of sensor clocking, blanking, etc. there could be a significant difference in the skew effect.

January 14th, 2008, 02:14 PM	#5
Eric Pascarelli Major Player Join Date: Nov 2007 Location: Los Angeles, CA Posts: 498	Noah - good point. I will have to try 720 - knowing what I know now I could easily to it on my fluid head, analyzing two successive frames, and won't need to use the moco. My guess is that it will be the same, but it's worth a try. Leonard - I will post a quicktime of the pan - I'm just sorting out with Chris H. why I can't upload a small mov file.

January 14th, 2008, 02:39 PM	#6
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	it seems that it would be difficult to guarantee that the pan speed will be constant each time without motion control. perhaps there is some object that can move across the screen a known speed? without knowing precisely the speed of the pan/movement it would be hard to be sure of the results especially if the differences are not large in relation to the size of the video frame.

January 14th, 2008, 02:44 PM	#7
Steven Thomas Trustee Join Date: Sep 2005 Location: Gilbert, AZ Posts: 1,896	Eric, thanks for this test. Can you give us an idea if it's close to the RED ONE numbers below for skew based an image moving across the sensor at a consant speed. Jim Arthurs' tests show with the RED ONE 18 frames will skew 3% of the 4K width. This means if shooting at 4K 24P, a constant speed pan (good luck ;) ) from one edge to the other takes 750ms, it can cause 3% skew on the 4K images. Jim also mentioned if the pan speed was reduced to 3 seconds, there's virtually no visable skew.

January 14th, 2008, 03:05 PM	#8
Steven Thomas Trustee Join Date: Sep 2005 Location: Gilbert, AZ Posts: 1,896	Eric, if you still have the test stuff around, can you do the same thing for tilt (up/down) movement. I'd like to see this measure affect too.

January 14th, 2008, 03:57 PM	#9
Vince Gaffney Regular Crew Join Date: Dec 2005 Location: Chicago, IL Posts: 146	So Eric, Is it safe to assume that since you were long on the lens and in a practical way took the arc out of the equation, we should assume the same results for a dolly - or tracking move? If all other factors are relative - frame, focal length, distance from object etc. vince

January 14th, 2008, 04:14 PM	#10
Eric Pascarelli Major Player Join Date: Nov 2007 Location: Los Angeles, CA Posts: 498	The point of going full tele was to take the arc and the rectilinear lens distortion out of the equation as much as possible. Even with the arc, the measurements were taken near the center of the image, and from the same part of frame, so the arc is not significant. Skew should be considered a 2D phenomenon that happens on the flat surface of the sensor - is doesn't matter how you arrive at the motion that's projected on the chip - it could be dolling, panning, zooming, subject motion or some combination thereof. But if there's something that should appear vertical moving across the chip horizontally it will be skewed. Likewise, as I'm sure is obvious, if you are panning or dollying with a moving subject, the subject will not be skewed and the background will.

January 14th, 2008, 07:24 PM	#12
Matthew Roddy Major Player Join Date: Jan 2006 Location: Riverside, Ca Posts: 307	I hope everyone will forgive a very novice question (I've never heard of skewing, so this is all educational for me): Is Skewing an inherent trait to all DV cameras? or CMOS only? For example, I've not noticed it in my Canon XL-H1, or my Sony Z1u. Which is to say, I've never looked for it and rarely do fast pans and the trait may very well exist. I look at the examples and am somewhat awed.

January 14th, 2008, 08:42 PM	#15
Mooho Bae Regular Crew Join Date: Jan 2005 Location: Korea Posts: 52	Thank you so much Eric. I'm interested in the 24P case. Is the scan speed still remained at 1/60 sec, or reduced to 1/48 sec. Would you please examine that?