[Spodeworld]

I noticed that the new Olympus camera (forgot the model) has 5-axis image stabilization. I wonder if Pentax, which uses a 2-axis approach, will move to implement a more sophisticated SR approach in near future as a result.

[RKKS08]

Pentax SR in fact is compensating in 3 axis; different from some of the competition, the 2-dimensional shift allows to correct rotation, too (within the same plane).
I am not sure whether the "5 axis" approach is more of a marketing argument, if it uses the sensor position to compensate. I don't think it would be a big improvement except perhaps for macros.

See also Falk Lumo's works on the subject:
Falk Lumo: Pentax shake reduction revisited

Care also for his links.

[Spodeworld]

It would be interesting to find out if the additional dimensions do buy you something. If so, it would be great to get a more effective SR.

[RKKS08]

I was a bit to quick with my answer.

Compared to the Pentax K5, Sony has added rotation around the x-axis and y-axis (the optical axis seen as z-axis), so SR could indeed improve a lot. I had thought only about z-shift, which would not give a significant improvement.

The camera is the mFT model OM-D. See the preview of Dpreview here (no test of the SR system yet):

Olympus OM-D E-M5 Preview: 1. Introduction: Digital Photography Review

[Spodeworld]

I hope Pentax takes this as a challenge and ups their SR game.

[RKKS08]

Added space in direction of the optical axis would be needed (it would probably add >3mm to the thickness of the body), and 2 more gyros would be needed. And the energy consumption of the SR system would increase by >40%. And a complete new system of fixing the grid array for the sensor within the body would have to be implemented (with great care not to touch any patents of Olympus).

[Lowell Goudge]

Since rotation is only a combination of x and y axis movements I am sure Pentax already has it in place. Additional sensors could add a completely different dimension, not rotation or translation of the z aziz but rotation of the z axis as opposed to simply lateral translation. Put the sensors at the mount, and combined with ones in the sensor plane, it allows to consider better how the vibration is taking place, this could really help on tele shots, but then again, since I have already pulled off lots of shots using a 500mm lens and both my K7 and K5 at 1/40 to 1/100 does it really need to get better?

[RioRico]

I'm anxious to see a 5-axis (5-dimension) SR system. I can grok 4 of those -- x,y,z, and t. What's the 5th axis? Luminosity? Tau+teh? Is the SR system formed as a hypercube?

[Lowell Goudge]

If you do a little research on other motion systems you can find refences to 5 and 6 axis flight simulators, a 6 axis simulator has roll, pitch, yawl, and then vertical horizontal and longitudinal. The use of the word axis or more accurately degrees of freedom described theb6 different motions. 5 axis would cover vertical horizontal, and rotational or roll ( the three easiest to put onto a sensor based SR) as well as pitch and yawl, which it means it can determine not just the longitudinal axis shifting up and down but also pivoting around some point, likely due to the body moving around the pivot point of your left hand.. It would require, as I posted earlier accelerometers on two planed, likely the lens base and sensor, and ideally also knowing the focal length of the lens

[Quicksand]

I'm anxious to see a 5-axis (5-dimension) SR system. I can grok 4 of those -- x,y,z, and t. What's the 5th axis? Luminosity? Tau+teh? Is the DR system formed as a hypercube?

Mmm hmm. If Olympus can deliver on an SR system that can move forward and backward in time, then FOR SURE I'm leaving Pentax. Perhaps I already have!

This is all such marketing silliness. Axes are not the same thing as degrees of freedom, and we're talking about the latter here. And this is for motion sensing only! Compensating for more than three kinds of movement (horizontal, vertical, rotational) at the sensor makes no sense. Pentax already does this.

And I'll add, the utility of translational motion sensors decreases greatly as the distance-to-subject increases. Maybe Olympus SR is OMGAWESOME!!! Or maybe they're engaging in some puffery.

Sigh.

[RKKS08]

First of all, prior to retirement I was working 25+ years with coordinate measurement systems - the first 8 years as an employee for a manufacturer of such systems, then as self-employed as an adviser for companies who use such systems. Whereas during the last years I earned my money by inventing and implementing solutions for tricky inspection problems in quality control, and writing software as a user friendly front end for such solutions (not all users are engineers), before I was also heavily involved in writing software to map and correct machine-inherent errors which cannot be fixed by hardware solutions. Since that time I may have forgotten a lot of the most complicated mathematics involved, but I still have the "built-in" feel to "think" in moving and rotating 3D-systems - a lot of people (and, for some reason, most women) cannot "think" in 3D geometries.

With this background in mind, I must correct some of the misunderstandings in the last posts, before they enter the public knowledge base and will be repeated in postings for the next 10 years:

@RioRico:

I'm anxious to see a 5-axis (5-dimension) SR system.

It is a 3-axis (3-dimension) SR system, using 5 of 6 possible parameters of the coordinate transition matrix. Pentax does use 3 parameters.
This link
Olympus OM-D E-M5 Preview: 1. Introduction: Digital Photography Review
offers a drawing which exactly shows what the Olympus system does.

@Quicksand:

"This is all such marketing silliness. Axes are not the same thing as degrees of freedom, and we're talking about the latter here. And this is for motion sensing only! Compensating for more than three kinds of movement (horizontal, vertical, rotational) at the sensor makes no sense. Pentax already does this."

This is just not true. To translate complicated matrix mathematics to real world language: you can shift along and rotate around every of the 3 axis, where shifting can be in 2 orthogonal directions. This would make 9 possible alterations of a coordinate system within another coordinate system. However, there would be redundant information embedded. In fact you can describe any such movement by 6 instead of 9 parameters. To translate for better imagination: you can shift in 3 orthogonal directions, and rotate around 3 orthogonal axis, so 6 transition parameters are involved. Olympus uses 5 of these, just ignoring shifting along the optical axis. And this missing one just effects focusing (distance to the object); that's why I said in a previous posting it may effect macro only.
The Pentax system does NOT compensate the 2 additional rotations shown in the drawing in the dpreview link. It shifts along 2 axis and rotates around 1 axis. This is more than some of the competition does (or did): when Pentax introduced the system, they were proud to announce their sensor also rotates around the optical axis, which many of the other systems (at that time?) didn't.

@Lowell Goudge (posting of 02:59):

"Since rotation is only a combination of x and y axis movements"

This is not true, neither in the mathematics involved, nor in the technical implementation to correct it. "Shifting" in mathematics as well as in natural language usually is a synonym for "parallel shifting" - if you hold a point of a device fixed and "shift", you really do "rotating" (in fact you are using a lever around the "fixed" point).
Of course you expressed very cautiously, using "move" instead of "shift". But then you said x-movement and y-movement - which, in my understanding, means shifting, excluding rotation.

"The use of the word axis or more accurately degrees of freedom"

- that statement is of course correct if you have a look at the underlying mathematical theory. And it is needed if you are dealing with multi-dimensional (>3) systems - which does not apply for mechanical problems. However, no one living outside mathematical theories (and physics/mechanical engineering) will understand it correctly. I prefer to use an "axis" model when explaining things to non-professionals.

[RioRico]

I hope all realize that my comment was facetious. In 3D space we can work with as many dimensions or axes as we wish, where each 'dimension' tracks a different variable, and each axis points in a different direction. I long ago wrote software for the old SoundBlasterPro card within 31-dimensional space. But n-axis movement still translates into a 3-axis model, as I understand it.

[Lowell Goudge]

RKKS08. Of course you are correct relitive to rotation. I was describing a point not a body. But really the discussion all relates to being able to determine the movement of the body relitive to the nodal point of the lens. This is what they are trying to get to. As a result they would need tis data also or as a minimum focal length and guess

[RKKS08]

"But really the discussion all relates to being able to determine the movement of the body relitive to the nodal point of the lens."

Again, this cannot be used as a definition. With the lens fixed on the body, the normal from the sensor to the nodal point never changes, whatever you do to move the camera. If you describe it by using the movement of this very normal in space, it would be a "cleaner" definition. But as a plane can be mathematically defined by a point within the plane and the vector described by the direction (its co sinus values) of any normal line on that plane, we are back at the usual definition: the position of the sensor plane in space (relative to the object).

Edit
Before anyone is now searching in books or Wikipedia for the definition of a plane: as for any geometrical object, there are many possible definitions for a plane (could be the coordinates of 3 points within the plane), and formulas to transform one into another do exist. The [point/normal direction] is preferred for coordinate transformations, because algorithms get much simpler.

[Quicksand]

This is just not true. To translate complicated matrix mathematics to real world language: you can shift along and rotate around every of the 3 axis, where shifting can be in 2 orthogonal directions.

It absolutely is true -- I stand by what I said. There's no reason whatsoever to move the sensor out of the plane of focus, which is what any movement beyond the three I described (horizontal translation, vertical translation, in-plane rotation) will do. It won't compensate for camera movement. It will only screw up your photos.

I agree it's POSSIBLE to move the sensor with six degrees of freedom. But you wouldn't want to do it, unless you want to defocus your image or induce faux-tilt-lens effects.