rolling shutter at high shutter speed?

[Stephen Henderson]

I was thinking the other day, shouldn't a high shutter speed like 1/2000th of a sec fix rolling shutter.
I know no one wants to use that, but might come in handy in certain cases.

Anyone know if that works?

[Daniel Epstein]

In theory while you think you are fixing or minimizing the rolling shutter issue by shooting at 1/2000th you are also creating issues by using such a fast shutter. For motion this becomes readily apparent even at shutter speeds above 1/125. Some people don't even like the look above 1/60th. For Stills this may not be as much of an issue. Also you will need a lot of light or a higher ISO which might add noise in order to shoot at 1/2000th.

[Tom Dickerson]

I read to minimize roller shutter you should shoot 60fps at 1/120th (no faster). Beyond that is to work within the limits of what you find acceptable.

[Evan Donn]

Shutter speed has no impact on 'rolling shutter' in these cameras because they refer to two different electronic processes (unlike a camera with a mechanical shutter). 'Rolling shutter' means the image is scanned sequentially, one line at a time, from the top to the bottom of the frame - and the delay from the top line to the bottom line is what causes skew & jello. The length of that delay is determined y how fast the processor can read and process lines. Shutter speed affects how long each individual line is exposed before it is read, but doesn't affect the speed at which the lines are read out. So even if you only expose each individual line for 1/2000 of a second it still takes somewhere around 1/60th of a second to scan from the top to the bottom line.

[Daniel Epstein]

Evan,
Yes they are two different processes but there should be an interaction. The "Rolling shutter effect" shows up when the read out time of the sensor is slower than the motion of the image being read from the sensor allowing the image to be displaced on the sensor over time. The skew and jello are not caused by the readout but are a consequence of the motion being read over time instead of simultaneous reading every line. If the faster shutter speeds are being done correctly then the image to be read out of the lines info on the chip would be based on the shutter speed when it is faster than the rolling shutter effect from the line by line read out. When the shutter speed is close to the read out speed or slower then the "rolling shutter effect" should be more visible. The still of an image in motion taken at a 1/2000 of a second shutter but read out at 1/60 of a sec should have less blur skew and roll than a 1/60 of a second shutter read out at 1/60th of a sec.

[Jon Fairhurst]

Daniel,

Evan is right. The rolling shutter is independent of the shutter speed. I've done tests on the sensor with a flash at various shutter speeds.

When a frame starts, the first line starts capturing light. At a set time afterward the next line starts capturing, and so on, and so on. The delay between the lines starting to capture doesn't change.

The shutter time varies how long each line is capturing light. With a 360 degree shutter, the first line would be capturing all the way until the next frame fires. Where it gets confusing is that the last line can be capturing well into the next frame. It's that rolling overlap that messes with our heads.

On the 5D2, the delay between the start of the first line until the start of the last is 25ms. (At least that was the case at 30p. I haven't done the tests since the 24p firmware came out.) The time was 25ms regardless of the shutter speed.

[Daniel Epstein]

Jon.
Of course the read out speed of the chip is independent of the Shutter speed. Iris is also independent of the shutter speed but they do interact to form the image we see. The read out speed might change based on the frame rate the camera is set to record at with the fastest readout being based on the the fastest frame rate the camera can record. Or they might have just set it at the fastest rate all the time. The question is the image being read out at faster shutter speeds captured at the faster shutter speed or is it being updated based on the read out speed. This would mean in effect the shutter speed is limited to the read out speed.
Stephen was talking about shutter speeds which were faster than the read out speed and whether that preempted or disguised the Rolling shutter effect. Given the other changes to the image which occur when using the faster shutter speeds this idea would seem to have limited usefulness.

[Jerry Porter]

Another key issue here is that with the shutter speed set that high with 30fps is you will get a severe stobing effect also. Think opening scene of Saving Private Ryan.

[James Donnelly]

The big debate! Who is right? Will this one run and run? Grab your popcorn folks.

I'm interested because I'm wondering about something. If confirmed that jello is reduced by higher shutter speeds, is it possible, in theory, that the strobing could be fixed in post?

[Daniel Epstein]

I don't think this is much of a debate as we all agree the artifacts are not what we would like. By the way by my math 25ms is 1/40th of a second 1000/25=40 which is a slow read out and gives you a clue why the "Rolling Shutter Effect" is an issue. Plenty of time for objects in motion to be in different positions as the lines are read. Also at that rate I would suppose the camera has to read the chip out faster if you want to shoot at 60FPS (Limited to 720P and SD) as 1/40th would not read full frames correctly so there should be less Rolling shutter effect although still not a very fast read out of chip. My guess is they end up with 1/80 at 720 60P. If they could read the lines faster at the slower frame rate than they might have a better result if it didn't cause too much trouble with the rest of the processing.

[Jon Fairhurst]

Trust me. With the 30p firmware, the last line captures its image 25ms after the first line. It doesn't matter if you're shooting at 1/30 (which we measured to be 1/33) or at extremely fast shutter speeds.

What this means is that if you have a telephone pole from top to bottom in an image is that if you pan so you go from edge to edge in 250ms, the telephone pole will be offset at the bottom by 10% of the screen. With a fast shutter, the pole will be crisp. With a slow shutter the pole will have motion blur. But the amount the the pole leans will be exactly the same.

To test this, put the camera on a tripod and aim it at a neutral or dark flat field and start recording. Using a fast, strobe flash, trigger it manually over and over. Find an image where the strobe is captured within a frame. Find a pair of back to back images where the strobe starts in one frame and ends in the next. By counting lines of those two situations, you can determine the shutter speed and the rolling shutter delay.

Shutter speed is related to time duration. Rolling shutter is related to time offset. The two are completely independent.

And keep in mind that the last line might be capturing light from the previous frame at the same time that the first line is capturing light from the current frame.

[Evan Donn]

Originally Posted by Daniel Epstein

The question is the image being read out at faster shutter speeds captured at the faster shutter speed or is it being updated based on the read out speed.

The image is always captured at the same read out speed regardless of the shutter speed - so the rolling shutter effect is constant. The read out speed is limited by the speed of the processor in the camera and is presumably running at the limits of the current hardware in order to minimize rolling shutter.

This would mean in effect the shutter speed is limited to the read out speed.

No, it wouldn't - I still think you don't understand how this works. The two are completely independent of one another and thus there is no interaction between the two. Shutter speed is simply a measure of how long each line on the sensor spends gathering light. It always takes 25ms to read from the top to the bottom of the frame, regardless of how long each individual line is exposed. If any movement occurs in that 25ms period it will cause skew. In fact, higher shutter speeds would be likely to cause a slight increase in perceived skew (although actual measured skew would be the same) during fast horizontal movement simply because there is less motion blur.

[James Donnelly]

Thanks Evan, that is the best explanation yet, and finally I understand it.

[Joel Perkins]

would it be possible for a cmos chip to scan random pixels instead of going down the frame line by line? let's say that the cmos sensor has 1920x1080 pixels. when it scans line 1, it's scanning 1920 pixels, line 2--another 1920, until it gets down to the 1080th line at the bottom. right? maybe not, but that's how i understand it.

so why can't the chip scan 1920 random pixels at a time, like maybe it scans 34 pixels of the first line, 13 of the second, 182 of the 19th line..i mean you could come up with some kind of pattern or something so that it scanned 1920 pixels 1080 times per frame but never did it the same way twice in a row.

that way, when you moved the camera from right to left really fast, instead of vertical things coming out diagonal, wouldn't they just look blurry? or maybe it would look really weird and cause a whole new set of problems.

i assume that if it were a good idea, it would already be happening, so what makes it impossible/impractical?

ps--i don't know how to start a new thread or whatever, which is why i'm asking on this post.

[Steve Phillipps]

I had exactly the same thoughts Joel, but like you assume that for whatever reason it won't work.
To start a new post go to the main DVInfo forum page The DV Info Net Forum, choose your category then click New Thread.
Steve