XY Resolution and Accuracy for Stitching

[ray_parkhurst]

I have built an XY table using THK KR15's to use for stitching and it has been working well, but I am not happy with the method of moving the subject. I'm thinking that moving the camera will be a better method for my vertical system, as it would eliminate the need to move the lights with the subject. It would also allow either moving the camera/lens together, or keeping both lens and subject fixed and moving the camera across the image circle. It's the latter method that is most intriguing to me, since it eliminates the need for the lens to be telecentric. It also has the effect of increasing the physical step sizes needed for the camera, since the movements are in magnified image space rather than object space.

This increase in step sizes has me asking if the method of using a high precision linear actuator like the THK is necessary for the much larger steps used for stitching. Even with my current system (object space stitching), my steps are in the >5mm range, and if I move to the image space method the steps are very large, just slightly smaller than the sensor dimensions. The question is: can a different method be used?

I'm looking at using direct stepper drive via drive belt, similar to what's used in 3D printers. I believe this method results in essentially zero backlash, and while the steps are larger than for the THK, they are not small versus the requirements.

So, what are the drawbacks? Is backlash near zero? How accurate and repeatable is this method? If it's good enough for 3D printing, it seems it should be good enough for stitching, but I am not sure of the limitations of drive belts.

Also, has anyone implemented this method for movements? I've seen drive belts used to couple motors to rails, but not for direct drive since the step sizes are too large for stacking.

[JohnyM]

Im 90% done in making XYZ horizontal rig that moves optical train and subject + light remains stationary. Gimme 3 weeks and i'll post details.

I ruled out stationary lens method due to lack of optics with big enough image circle. Instead im interested in comparing so called "super resolution" vs panoramic stiching.

[Lou Jost]

I use the method of scanning over a single image, and I like it. I do it with MFT. If the stitching is not too large, a Fotodiox shift adapter lets me do it without any special x-y devices.

For APS or FF (and MFT too) I have adapters for view camera backs and I can shift using the geared x-y movements of the rear standard. With this you can construct quite large mosaics. I have not tried it yet. For this I would probably stick with moving the subject in the z-direction because of the weight of the system, but you have heavy-duty rails that could handle the weight.

[Lou Jost]

I should add that I am unsure whether this method eliminates the need for telecentric lenses. Since stacking algorithms have to scale and align each frame, it is possible that the scaling and alignment might not be exactly the same in each column. I have not done this enough to know whether this is a real issue or not. At the very least, each column would need to start in the same plane on the subject. So you would need a controller and rail that reliably returned to its original location after each stack. I can do this with the WeMacro controller on a MacroMate moving a focusing block.

[ray_parkhurst]

I should add that I am unsure whether this method eliminates the need for telecentric lenses. Since stacking algorithms have to scale and align each frame, it is possible that the scaling and alignment might not be exactly the same in each column. I have not done this enough to know whether this is a real issue or not. At the very least, each column would need to start in the same plane on the subject. So you would need a controller and rail that reliably returned to its original location after each stack. I can do this with the WeMacro controller on a MacroMate moving a focusing block.

I would turn off any scaling (I do this routinely anyway) to avoid aberrations created by the stacking algorithm. Alignment should not be an issue, since each "column" must be aligned with the others anyway. Note that my subjects of interest will be coins, so the stack is not very deep, typically <500um.

The panorama I'm looking at is likely 4 wide by 6 tall, so quite the task if done manually.

[ray_parkhurst]

I use the method of scanning over a single image, and I like it. I do it with MFT. If the stitching is not too large, a Fotodiox shift adapter lets me do it without any special x-y devices.

For APS or FF (and MFT too) I have adapters for view camera backs and I can shift using the geared x-y movements of the rear standard. With this you can construct quite large mosaics. I have not tried it yet. For this I would probably stick with moving the subject in the z-direction because of the weight of the system, but you have heavy-duty rails that could handle the weight.

I would plan to move the subject in Z. I actually don't like this idea, but I think moving the Camera + Lens + XY system would be an order of magnitude more difficult and problematic.

I can move the coin in Z with a mjkzz VCM system, but I would prefer to use a stepper-based Z mover since I'm using the mjkzz SnS system and the VCM does not integrate with it, so I'd need to launch each stack manually. Not a showstopper but not ideal. I have a small Z-stage I built from a precision-grade SKR2001A with a Lin 417 stepper, so that's my backstop.

[ray_parkhurst]

Im 90% done in making XYZ horizontal rig that moves optical train and subject + light remains stationary. Gimme 3 weeks and i'll post details.

I ruled out stationary lens method due to lack of optics with big enough image circle. Instead im interested in comparing so called "super resolution" vs panoramic stiching.

I tried the SR method, and it took me as far as I have ever gotten with single images, but the pano stitching method beats it by a wide margin. I initially considered doing both, but the problem is that for SR to work the images can't be diffraction-limited. There must be information present at sub-pixel level, or all you will get is a bigger copy of the same image.

[Lou Jost]

Super-resolution is wonderful but is limited to about twice the normal resolution. The resolution of a stitched image is limited only by your patience (and, if you are imaging live specimens, the amount of time before they grow or wither) .

[mawyatt]

The resolution of a stitched image is limited only by your patience

Lou,

My patience is limited, so the very reason for developing a completely automated S&S System

I did a few S&S sessions manually, that convinced me to look for a better way

Best,

[Lou Jost]

Indeed, a fully automated system frees you from almost any limits (as long as your targets are dead). I greatly admire your solution. I don't have your electronic skills and my targets are often not very stable, so I have to find shortcuts, and those come with limitations. But I can live with those. Usually I only need a little bit of stitching, most often just a single row. If one can get by with just one or two rows, everything simplifies immensely.

[mawyatt]

Lou,

One thing I've tried to do is create a system that can be easily adapted, electronically and mechanically.

For example: I can mount a KR20 on a KR26 which is mounted to the vertical bar on my setup. The KR26 does the heavy lifting for Z axis focus stacking, while the KR20 moves the camera lens in the X axis (sideways) and another rail can move the subject in the Y axis. This configuration is nice for subjects that have a high aspect ratio, so the pano only has 2 or 3 steps in the Y axis, and many more in the X axis.

Best,

[mawyatt]

Lou,

Another thing to add, is with an automated setup you can cover the subject in much less time since everything is already predetermined.

With some of the new controllers the rail movement is so smooth that you might be able to almost "shoot on the run" since you are not exciting the resonate structures of the setup and thus causing the induced movements.

I had mentioned a new controller that has a special mode which spreads the impact of the motor drive current pulses and controls the speed profiles in such a way as to not excite these resonances. So using a camera with fully electronic curtains, I believe one could almost shoot without any appreciable standstill delay, maybe just long enough for the exposure with a little extra and if flash is used just the delay needed for the flash to recharge.

Anyway, this is where I'm heading

Best,

[Lou Jost]

That sounds very exciting, Mike!

[ray_parkhurst]

That does sound cool. What does the "on the run" enable, just more speed in the stacks, or something else you're thinking of?

I'm sort of heading the opposite direction though, so coming back to the original question, does anyone have experience with belt-based drives like are used in 3D printers?

I'm looking at 16T / 20T 6mm wide 2mm pitch pulleys, which with a 400step motor would give 80um / 100um full steps, or 20um /25um quarter steps respectively. I'm wondering how accurate and repeatable those steps are?

[mawyatt]

That (on the run) means to just move thru the focus stack as fast as possible with little delay to wait for induced vibrations to settle. I'm thinking of sub-second per image capture at 5X without any major image degradation due to induced vibration from the stepper/rail, and fully electronic curtain camera operation.

The system can be modeled as a Linear Time Invariant System if small steps are considered, and thus analyzed and optimized as such. This is the approach I'm taking and assumes many different things must come together and work well, so all pure speculation now (well almost all )

Can't help with any info on belts, other than they would "seem" to be more prone to errors in position due to stretching, binding, aging, and temperature effects.

Best,