"Rolling shutter" effect in video cinematography

[Doug Kerr]

I am somewhat hampered in this area by my lack of familiarity with the details of modern video camera operation - how the sensor is read out, how it is "reset" between frames, and so forth, for various sensor types.

I see mention of the "rolling shutter" effect, which seems to describe a phenomenon pertinent to video cameras that somewhat corresponds, on a "dynamic" basis, to the "image skew" effect we get in still photography with a focal plane shutter.

CanonRumors has posted a video piece by cinematographer John Yi, extolling the new canon C300 cinema video camera. In a couple of places, he alludes to a problem with video done with video-capable "still" dSLRs, when the camera moves rapidly. Because of download problems, it is hard for me at exactly appreciate the effect.

But I discussed it last night with Will Thompson, who thinks that what is spoken of is the phenomenon I mention at the outset.

And evidently the scheme used in the C300 averts or at least mitigates this problem.

I'd appreciate hearing from those who have some insight into this.

On a side issue, despite my always having thought of myself as well informed in broadcast analog television technology, I recently realize that I am not sure, with an image orthicon (perhaps the most-widely used pickup device for analog TV during its "golden era") whether (a) the image is frozen for each field and then then read out by the scanning beam or whether (b) the image is continuously "developed" on the target and scanned "in real time" (each element of each line representing its part of the scene at successively-later instants in time. I have always assumed the latter.

If so, then probably the phenomenon of which I speak would have occurred there as well.

Again, I'd enjoy some enlightenment on this from those who really know.

Thanks.

Best regards,

Doug

 

[Mark Hampton]

I am somewhat hampered in this area by my lack of familiarity with the details of modern video camera operation - how the sensor is read out, how it is "reset" between frames, and so forth, for various sensor types.

I see mention of the "rolling shutter" effect, which seems to describe a phenomenon pertinent to video cameras that somewhat corresponds, on a "dynamic" basis, to the "image skew" effect we get in still photography with a focal plane shutter.

CanonRumors has posted a video piece by cinematographer John Yi, extolling the new canon C300 cinema video camera. In a couple of places, he alludes to a problem with video done with video-capable "still" dSLRs, when the camera moves rapidly. Because of download problems, it is hard for me at exactly appreciate the effect.

But I discussed it last night with Will Thompson, who thinks that what is spoken of is the phenomenon I mention at the outset.

And evidently the scheme used in the C300 averts or at least mitigates this problem.

I'd appreciate hearing from those who have some insight into this.

On a side issue, despite my always having thought of myself as well informed in broadcast analog television technology, I recently realize that I am not sure, with an image orthicon (perhaps the most-widely used pickup device for analog TV during its "golden era") whether (a) the image is frozen for each field and then then read out by the scanning beam or whether (b) the image is continuously "developed" on the target and scanned "in real time" (each element of each line representing its part of the scene at successively-later instants in time. I have always assumed the latter.

If so, then probably the phenomenon of which I speak would have occurred there as well.

Again, I'd enjoy some enlightenment on this from those who really know.

Thanks.

Best regards,

Doug

doug,

rolling shuter - ffs check this oot about 1,30 in to the song - bloody annoyoing ...

help me git rid of this !!

http://youtu.be/2NP0uBTtKTM

ok ! git your reading gecks on man !

cheers

 

[Terry Lee]

I am somewhat hampered in this area by my lack of familiarity with the details of modern video camera operation - how the sensor is read out, how it is "reset" between frames, and so forth, for various sensor types.

I see mention of the "rolling shutter" effect, which seems to describe a phenomenon pertinent to video cameras that somewhat corresponds, on a "dynamic" basis, to the "image skew" effect we get in still photography with a focal plane shutter.

CanonRumors has posted a video piece by cinematographer John Yi, extolling the new canon C300 cinema video camera. In a couple of places, he alludes to a problem with video done with video-capable "still" dSLRs, when the camera moves rapidly. Because of download problems, it is hard for me at exactly appreciate the effect.

But I discussed it last night with Will Thompson, who thinks that what is spoken of is the phenomenon I mention at the outset.

And evidently the scheme used in the C300 averts or at least mitigates this problem.

I'd appreciate hearing from those who have some insight into this.

On a side issue, despite my always having thought of myself as well informed in broadcast analog television technology, I recently realize that I am not sure, with an image orthicon (perhaps the most-widely used pickup device for analog TV during its "golden era") whether (a) the image is frozen for each field and then then read out by the scanning beam or whether (b) the image is continuously "developed" on the target and scanned "in real time" (each element of each line representing its part of the scene at successively-later instants in time. I have always assumed the latter.

If so, then probably the phenomenon of which I speak would have occurred there as well.

Again, I'd enjoy some enlightenment on this from those who really know.

Thanks.

Best regards,

Doug

You are correct, it is the identical problem as the image skew..it is a scanning problem.

With the chips on the camera, as I understand the software, it is a single pass (at least on my Cannon 7D) for every frame. So they read the matrix starting with top row 0 down to row 1080 (I understand there is some interpolation happening as well as there are more than 1080 rows on the chip, but for problem purpose it is a single scan line for every frame). This scan is happening at 1/60th of a second (well, 1/50th is the closest setting you have on the camera) so 1/50/1080 seconds per line. Also,the chip doing the capture is always on, in other words it isn't like film, capturing 1/60th of a second then doing a scan of the latent image. As the scanning is happening, whatever light is hitting the sensor is being scanned at that moment as the software reads the values on the chip.

So with a fast pan a vertical line like the side of the building might be at column 42 on the matrix at row 1 of the scan, but by row 1000 of the scan it could well be at column 142. The result is very interesting to say the least.

Now for the old school:
With NTSC at least you had fields, 2 of them, offset one line at a time. And yes there COULD be rolling but the fields help to mask the issue. The "bad" rolling portion happened between fields for the most part as a scan was only half the image, top to bottom, at a pass. If you captured a full frame of fast panning you could see it, but it was between line1 and 2, line 3 and 4. The fields cut the problem in half. It looked more jagged, so the fields basically "interpolated" a closer looking straight line to the eye. So the offset or "rolling shutter" wasn't as noticeable in part because of the scan lines.

Plus the tubes in the camera and the TV were great at phosphoresing (is that a word? glowing maybe better) for milliseconds or longer, helping to mask the issue, again providing an analog form of interpolation. Even early HD 1080i helps mask the issue with fields scan lines.

PLUS with SD you had only 240 lines per scan in the same time frame, much less picture to worry about anyway.

The very first HD cameras were horrid, they had too much detail. We had to put software on them to tone them down!

The LCD TV, especially LED based super contrast, with the increase in line counts are part to blame for the problem. The software/chip design is the other design problem. My guess is the engineers that put video in the modern DSLR were not video guys, and didn't think about rolling shutter...while the basic design is awesome (I LOVE my 7d for video work) it is not really a video camera design.

I have no doubt the chip can "hold" values, as you can shoot the 7D at high shutter speeds, faster than what the software can probably read the chip. But in video mode the mirror is locked up, and the chip is always getting light. The scanning of the values is the issue. Faster parallel processors is what would be needed or multiple sensors, which is what you find on high end video cameras.

You can do several tricks to "fix" the problem on the camera in software to make it somewhat better, but I expect manufacturers to keep releasing new and better cameras instead, as what is really needed is different hardware :-(

Best piece of advice I ever got "If it hurts when you bend that direction, don't bend that direction!" With a DSLR just don't do fast pans and you are golden. If your script requires it, use a real camera for video capture!

 

[Doug Kerr]

Hi, Terry,

You are correct, it is the identical problem as the image skew..it is a scanning problem.

Thanks for the update and additional insight.

Yes I'm old-school (first studied NTSC in college - still have my textbooks!).

Best regards,

Doug