some real-world extreme shake: raw footage @35mbps
 

referencing this thread: http://dvinfo.net/conf/showthread.php?t=109548
that was a lab-condition-like extreme sample: at full-zoom with convulsive shaking.

here is a real-world example of extreme shaking. i put it in the same directory i used in the other thread:

http://bakraa.com/ex1supershakey

NOTE: these two file are 88MB please right-click and download (to avoid killing my server). it would be great if someone would volunteer to mirror them. thanks!
424_0096_01.MP4 - 720p24@35mbps (19seconds)
424_0096_01.mxf - same as above but in mxf wrapper

wildmotion.jpg - still from clips above showing little artifacting
wildmotion_CCd.jpg - same still CC'd radically (still little artifacting)

here i'm just moving the camera around wildly while climbing up a hill. i think the lens was at or near its widest.

to be more useful i've posted the actual .mp4 and corresponding converted .mxf files so you can also judge the codec, since shaky shots of bushes are difficult to encode. a similar test in the exact same spot with an HDV camera caused the codec to break up obviously and objectionably.

i think i shot this at 1/120th, at about f2.4, cinegamma1, colormatrix=cinema, sharpening off. manual everything, i think except for focus. no gain.

the rolling shutter distortion is less apparent in this intentionally shaken real-world footage than in the laboratory-test-like footage in the other thread. it's my short experience that rolling shutter distortion is barely or not apparent in actual footage.

Thanks Ali for posting these.
Comparing the new .mxf hill climbing-footage with your laboratory-test-like footage in the other thread I too find the rolling shutter effect almost invisible - good to know!
And yes, the codec quality (bushes) looks good compared to HDV.

Thanks Ali

Thank you, Ali. Compression wise, I am impressed. As for rolling shutter, I could not see anything of concern during playing back at normal speed via VLC media player on my computer. If I play back via VLC at maximum "play slower", I can then see some minor result, and only then.

I am satisfied that rolling shutter should not be an issue for my intended use of the EX1. If I was concerned for certain shots, I would consider shooting my particular setup of a 24p (or 25p) shoot, such as an intense hand held running scene (think: 28 Weeks Later) as 48p (or 50p), then in post playout at two times (2x) speed playback, effectively dropping every other frame and it should fit into 24p (or 25p, respectively) nicely. So instead of each frame taking 1/24 of a second to capture the image, it will only take 1/48 (or 1/50) of a second per frame, reducing the prospect of rolling shutter, and then drop every second frame in post. This should help reduce the rolling shutter issue by increasing the frame record speed, per frame. That is the theory....

Please confirm, would you need to record audio to a separate external device, as certain off speed modes do not record audio at all to the EX1?

I assume that Sony have minimised the rolling shutter effect by always getting the chips to read as fast as possible, so I would be surprised if increasing the frame rate makes any difference to read speed.

Audio is not recorded at any non-standard frame rate. 720/50p (or 720/60p) would give you audio, but you would have to lift it off before dropping frames, then drop it back in and sync it.

Nick

Yes Nick, I heard a description of that very thing at DV Expo West from Juan Martinez, Senior Manager of Technology, Sony Electronics. I wasn't in for the entire presentation, but he said that the Exmor CMOS chips are very fast, thus eliminating that 'rolling shutter' effect.

Even if this should work wouldn't it be much easier to just increase shutter speed? Or am I missing something?

All the more reason for me to shoot 1080p25, as 720p50 would record audio, that is handy for us 25p/50i countries.

As for my suggestion of shooting 48p, I understand your suggestion Nick and James, and it may end up being correct. Perhaps a real world test in due course will be conducted, pending specific clarification of how EXMOR functions in this comparative proposition.

Ola, my understanding is shutter speed will not vary the speed of each row being recorded on a rolling shutter system.

anecdotally, i think the rolling shutter effect is smaller at higher frame rates. i remember looking at some of my shakey 60p footage and noticing it had surprisingly less distortion. will have to check again.

i'm also guessing that some wholesale reclocking is occuring when you change the frame rate, becuase the lcd screen blanks out and red scan lines jump across the screen for a second (much the same way your computer lcd display does when changing the scan rate). the clocking tree and buffering scheme for the sensor would get more complicated if it required an odd divisor (like 60/24->5/3). i'ts a good guess that the rolling shutter scan rate is based on the frame rate and is not always clipping away at the fastest it can go.

also the slower the shutter, i'd bet the less distortion you get and not just because you get more motion blur. but because more of the frame is getting exposed simultaneously before getting reset... so the part that is skewed is less exposed than the part that isn't. if you imagine that the rolling shutter is like an electronic slit moving top to bottom: so rows get reset in turn, scanning down the image; then rows behind it get read. there is a fixed distance between the two, like a slit. the best case is that the reset scan happens REALLY fast, the whole image is exposed together, then the readout scan happens REALLY fast. that would almost be a global shutter. we know this not to be the case. we also know the scan can take no longer than the reciprocal of the frame rate (it looks like it's faster than this, but would have to do some checks with a calibrated strobe light or somethign similar). the worst case is that the slit is small and the scan takes the reciprocal of the frame rate.

it's another reasonable guess that the individual row scans happen in parallel or fast enough to be considered parallel: if you were to support say 1/1000th of a second shutter you'd have to clock out the rows pretty fast. so say this is the case, then if a scan at 24fps takes 1/24th of a second and you need to support a 1/48th shutter speed. then the "slit," or the distance between the row reset and read has to be 50% of the total height of the sensor, or 540 rows. to get 1/24th shutter speed, you just reset right after you read, so it's effectively no shutter.

i have a feeling that if you shot at 50p with a 1/60th shutter or maybe no shutter, and threw away every other frame, then slowed everythign down 4% (for 24p) you'd get about half as less rolling shutter warping at the expense of slightly jerkier movement (skipping every other frame).

This test proves at least one thing... it is a bit faster, but not fast enough for extreme documentary style shooting. That's where XDCAM HD gets in.

No doubt the F350 is by far the better camera, but I would consider the doc that Jody Eldred did with the planes to be on the extreme side. At least for me. LOL