Image stabilisation in camera bodies

In-body stabilization just adds another mechanical moving part that could fail.

dbdors wrote in post #6032344
Why is that such a big advantage. I don't care whether I can see it or not, I just want it to work. And it works quite well.

Well, I certainly DO care about being able to see the effect. A stable viewfinder image is critical in many situations. For a couple of examples:

1. Manual focusing is all but impossible with long lenses, if the image is wobbling about. It also usually wobbles a little more when focusing, than when in a solid "braced and steady" shooting stance, due to the hand on the lens moving the focus ring rather than just holding the lens steady.

2. In motorsports or aircraft photography, when you are using slow shutter speeds to blur backgrounds or props, the subject will move a distance whilst the shutter is open. To get an accurate pan, most people will place a focus point over a specific part of the subject (pilot, door mirror, whatever) and attempt to keep it bang on that position as they pan. The more it wobbles around due to camera shake, the harder it is to keep the point on target. In effect, the in camera IS may be keeping it on target, but as you can't see that you will be compensating manually for the movements you are seeing in the viewfinder, in addition to the IS compensating, therefore introducing shake in a situation where the slow shutter speeds are already making it hard enough to control.

To be honest though, I would probably turn off in-camera IS in such a situation, as it has little effect with long lenses, has no mode 2 for panning (although that should be possible) and is likely to make things worse, rather than better.

On low end bodies, I can see the use of in-camera IS as they are mostly used with shorter lenses for family snapshots and it will work reasonably in that situation. Higher end bodies are used more by enthusiasts who are likely to have longer lenses in their bag, so will be better served by in-lens systems matched to each lens.

I would like to offer my thoughts from a control system engineer's point of view ... both types of image stabilization systems work, but the questions are how well for all lenses and at what cost. I won't address the cost since that is outside my expertise.

First, I should explain what control systems do for the benefit of those of us who are not technical types. In technical terminology this would be referred to as a "closed-loop feedback control system". In layman's language this means that some type of sensor is used to detect an error in whatever we are wanting to control and then from that error, produce an appropriate output that will drive the system to correct the error. During this process, the sensor is continuously measuring the magnitude of the error as the correction is being made and then from that revised error provides an update to the mechanism that is making the corrections until the amount of error reaches a null point(essentially zero or very close to zero).

In the case of using a closed-loop feedback control system on a camera to stabilize the image when using in-the-lens control, the sensors being used are accelerometers. Accelerometers measure acceleration which is a change in velocity. A smooth panning velocity might not be a problem if you are doing BIF or sports shooting, but the shaking and jerking part of the motion would definitely be a problem because it has much higher frequency components that would show up as image blur if the shutter speed is not extremely fast.

It is also possible to derive velocity from an accelerometer by taking the time integral of it output (if you are familiar with calculus) using the microprocessor in the lens. I do not know if Canon tries to stabilize for velocity which is a low frequency term, but even if not, the velocity term would be useful to provide control system stabilization by adding some damping.

Since I mentioned stabilization, that is one of the key parts in the design of an optimally performing control system -- you want to make sure that it is tightly coupled (meaning that the error is kept very close to its null point at all times) and at the same time both stable and responsive. The goals are minimal overshoot (sometimes called ringing) and minimal lag (sluggishness in response to correcting an error which will lead to larger errors). I also believe that stabilization is the area where in-lens control has an advantage over in-camera stabilization. In a mechanical control system such as this, things that have mass, and therefore inertia, (internal lens elements int this case) have to be moved by some type of control motor and drive train, which also has inertia. Dealing with inertia in a system means that the designer has to consider sluggishness in response to null an error and also overshooting a correction if if gain is too high or phase delay is too long. A control system is likely to exhibit both problems at the same time. Since the inertia characteristics of each lens is unique, the control system stabilization design for each size lens is "tuned" for optimal performance on each particular lens.

Finally, with an image stabilization in the lens, precise control is comparatively easy to do because the controller does not need to deal with much of an inertial load for the lens tilt mechanization that Canon uses. Additionally, a considerable amount of control authority is possible without needing to apply a large degree of lens tilt in stabilization. And because the mechanization only needs small amounts of movement to make corrections, the system can have very high bandwidth and at the same time be very stable.

Without knowing if the in-camera systems use accelerometers or some other means for detecting vibration and jerk (in calculus terminology, jerk would be the first time derivative of acceleration), but I think that it is sufficient to look at what would need to be done in the camera to counter the problem. Unless there is are corrective tilt lenses in the body, correction would involve moving the sensor itself both horizontally and vertically. I can envision that for a long lens that the amount of shake could be a significant part of the viewing area. So tracking in such a situation would involve moving a much greater mass over a very wide range while maintaining tight tracking response, keeping the sensor at the focal plane without any front to back movement, and avoiding compliance problems (springiness, looseness, and positioning accuracy) between the sensor and body would be considerable design challenges. I suspect that some non-trivial design compromises in performance would be necessary for an in-camera system. Even with in-camera image shifting lenses, the control authority (the amount of image motion that could be corrected) would require a greater magnitude of adjustment when done so close to the focal plane.

^Ha, ha, it's been awhile since I've read an engineering publication! Too bad I soon have to go back to school and finish my last semester. *sigh*

JouerCanon wrote in post #6034043
^Ha, ha, it's been awhile since I've read an engineering publication! Too bad I soon have to go back to school and finish my last semester. *sigh*

Hi, welcome to POTN. I was drafted during the Vietnam war when I only had three semesters to go before graduating. Shifting gears when getting back into the academic world took a bit while I brushed off the cobwebs from what I had learned prior to going into the Army. My major professor was understanding and told me that I could informally audit any prior classes without having to pay the extra tuition. The GI Bill was nice, but back then tuition was almost free at state universities.

Thanks! Yeah, I find it funny how times have changed. I know I haven't lived those times, but hearing how cheap college was and how differently people thought certainly makes me wonder how much society has changed. Right now tuition at Hofstra University (in Long Island, NY), is about $24,000 to $32,000 per year (two semesters). Thankfully enough ROTC pays for the bill and then some.

bill boehme wrote in post #6033581
I would like to offer my thoughts from a control system engineer's point of view
(snip)

Excellent synopsis, Bill!

-js

Thanks for the input guys. You've mostly confirmed what I thought, but you've fleshed it out a bit. I think a key thing is being able to see what you're taking. Although I have IS on my 24-105 I could well manage without it - it just enables me to be a bit sloppier when I use that lens. With the 100-400 IS really comes into its own. For many years I used a Tamron SP 70-210 on my Canon A1, and even at 210 I could get pretty good handheld shots down to about 1/30 second. Obviously I was using bracing techniques, which meant it was only useful for static subjects. With the modern lens I can take birds on the wing and consistently get pretty good results, something I couldn't have contemplated before. So far as I'm concerned, IS/VR is needed at the long focal lengths when any in-body IS system becomes ineffective.

With the increasing numbers of P&S cameras with IS, and there are several made by Canon, how is that achieved? How about the much-vaunted G9 for example? Is it within the lens or does the sensor move? Or is there some third method only applicable when the lens isn't removable?

Bill, thanks for the great write-up. That was some of the better reading I've done here on POTN.

I have reflected on this thread, as it seems quite well answered and almost left to the test of time. However, I have just decided to post my thought on the OP.

Two things that come to mind:

- Just like Self-Cleaning Sensors have been developed and eventually assigned to most camera models (I am not sure Full Frames have it though, such as 1DS3), I believe they will be out there for most cameras in the future. With IS (as related to the camera body), I do believe that both big players in the market will end up adding them to a few select prosumer bodies in future releases. I do however feel pretty confident that this will NOT carry over to the Professional Cameras. How so? Well, here comes the second point.

- The IS issue (In my humble and flawed experience) is mostly a principle of applied "Optical" principles, rather than a mere printed Circuit Board trick. Given the need for top of the line cameras to provide their absolute possible performance, IS systems would not optimize the result of imaging. Canon and the Digital world has helped me realize even more so now, that GLASS is king. Specially at 200mm and beyond.

So, in conclusion: Wanting an IS system in the camera body, is the same as asking for a 300mm lens that is only 1.5 inches long. Is it possible? I am sure some Tele Point and Shoot out there must have some pretty crazy zoom ranges without a 12-15 inch tube sticking out of the body. However, when the time comes to compare results, the response should be easily accepted. Can IS be implemented in Cameras? Yes, and I am sure Canon and Nikon will do it. But the Pro Series will always remain untouched, simply because of optimal optic performance. And at that realm, a 400mm with IS is not that much more expensive than the same lens, without IS.

At any rate, my view.

Canon have themselves stated that it may be possible to make an in-camera IS system capable enough in the future, but for the moment, they put it in the lens, as the result is superior.

And as I've stated, and Bill's nice document implies, stabilizing shake in a lens with a long focal length, and thus with a large magnification, isn't physically possible, as you have to move the sensor outside of the camera body.

apersson850 wrote in post #6036585
......... and Bill's nice document implies .......

Wow! My post has achieved document status.

JouerCanon wrote in post #6034708
Thanks! Yeah, I find it funny how times have changed. I know I haven't lived those times, but hearing how cheap college was and how differently people thought certainly makes me wonder how much society has changed. Right now tuition at Hofstra University (in Long Island, NY), is about $24,000 to $32,000 per year (two semesters). Thankfully enough ROTC pays for the bill and then some.

Tuition in Texas was $50 per semester in 1965.

Would the effects of in-lens AND in-camera IS stack if such a thing were ever to happen? I was thinking maybe yes, because the lens is trying its hardest to produce a straight path of light to the sensor, and then the sensor is moving itself as well to counteract any wobble. But then I thought about the feedback loop issues IS has...anyone know the anser?

Pandya wrote in post #6038162
Would the effects of in-lens AND in-camera IS stack if such a thing were ever to happen? I was thinking maybe yes, because the lens is trying its hardest to produce a straight path of light to the sensor, and then the sensor is moving itself as well to counteract any wobble. But then I thought about the feedback loop issues IS has...anyone know the anser?

I am certain that the camera would explode in your hands, more than likely taking at least one eye out......

Or the camera would turn against you. Conflicting missions.....
http://en.wikipedia.or​g/wiki/HAL_9000