Anatomy, A DIY Precision S & S System (Mechanical)

[mawyatt]

Hello,

This is the follow up hinted in this and other previous posts.

http://www.photomacrography.net/forum/v ... ack+stitch

Some brief background on the development of a DIY S&S System. Having engaged in manual S&S efforts some time ago and experiencing the painfully slow and tedious process (I don't have much patience) I decide to try and expensive automated system from MJKZZ. This is a nice system with good components but only operates under Windows, and I'm a Mac user. Hoping for a Mac version which hasn't happened (Peter never said he was going to do a Mac version, so my bad!) I ended up still doing the S&S efforts manually, and thought about developing a DIY Precision S&S setup without having to deal with Windows.

One thing I noticed about the MJKZZ setup was the horizontal configuration with a "hanging" portion which hangs off the end of a table, that was going to be a problem in some cases, so I though of using my horizontal Thor based setup for starters. Realizing the best possible solution would let the camera/lens move vertically and the subject move Y and X. I did some research and ordered a Thor 95mm 4 sided precision optical extruded rail, my horizontal Thor system is based upon the one sided 95mm version securely bolted to a large Thor optical base plate, so everything is compatible. The Thor sliding clamps & 95mm rails are amazing, more secure than anything I've every used.

I've drilled and tapped a few of the Thor clamps so that I can mount just about any focus rail, ARCA clamps & plates that I have. This includes Stackshot, Wemacro, THK KR20, 26 & 15, MJKZZ focus rails.

First off for the X and Y subject positioners I thought of just mounting a Wemacro or MJKZZ on top of each other at right angles using the ARCA clamps, this works and is easy to do, however is takes up considerable space from the base mount to the subject plane. For moving large distances with large subjects this works fine, but I wanted something smaller for smaller subjects like the chips I often shoot.

I had been watching ebay for awhile for a reasonable THK KR15, and when Ray spotted a good deal I jumped and purchased 2 of the US Automated KR15 rails. They are configured for an older RS485 interface, and I thought of getting a RS485 to USB adapter to use, but decided this was more of a "patch" than a good solution. So I removed the 2 electronics boards inside the rear case as shown, these are very well done BTW. Then I soldered the 4 motor wires to a small eBay connector that I glued to the top of the motor mount as shown. After some research the motors are NEMA 11 with 5.6 ohm winding resistance, 4.2mH inductance and 670ma current rating I believe. Caution, these motors will get very hot (P = two times I^2R) if run at high currents because of the high winding resistance, and I've found out after some experimenting they do not behave well at lower voltages with high currents. Keep the current below 670ma and use a 12 Volt supply.



The USA THK KR15 have adapter plates on the stage that can be drilled and tapped with a 1/4-20 thread which is standard for photo equipment, this allows various items like the small ball head to be easily mounted using a male 1/4 to 1/4 adapter as shown, note the 2 short curved head M3 bolts. These are used to mount one KR15 on top another using the 14mm spaced recessed holes in the base. Caution, do not use regular head M3 bolts, the heads will damage the carriage. The 4 longer M3 bolts are also curved head and are used to mount the bottom KR15 to an long ARCA clamp as shown that has been drilled and tapped for the 4 KR15 mounting holes. Caution do not attempt to enlarge the KR15 mounting holes, they are recessed and the thickness at the hole base is just a few mm. Just mount with short head M3 bolts and everything will work and have proper clearance.




The two KR15s are mounted one on top the other using the mention 2 (top) and 4 (bottom) M3 bolts as shown below.




For the vertical axis I use a KR20 with MKJZZ motor kit that is mounted to a 150mm ARCA plate, the bolts are a little long, but that's what I had, and there's nothing mounting or sliding inside the ARCA plate, so no risk of damage.




I often use the Wemacro Vertical Stand for vertical and horizontal use at low magnifications and this makes a great means for a small portable S&S Setup. I use a plate (Wemacro) that mounts to the stand base with 4 1/4-20 bolts (this is also sometimes used with stiff springs between the plate bottom and base surface for micro-leveling of the plate). The plate has been drilled and tapped for 4 M5 bolts that hold the 150mm ARCA plate as shown, drill the ARCA plate and Wemacro plate at the same time to assure precise alignment.


Here's how things hooked together mechanically for the Vertical and Horizontal setups. The horizontal setup uses an extra 1/4 to M8 adapter (Wemacro) to hold the Misumi 40mm square bar more securely, the Misumi is a longer (600mm) version of the supplied 40mm bar. This must be drilled for the long 1/4-20 bolts that hold the ARCA clamp to the bar for supporting the focus rail & camera/lens assembly. Use the Wemacro supplied 40mm bar as a guide for the holes in the Misumi 40mm bar.








Now, I'll start another thread with the electronics part and the final configuration.

Best,

[ray_parkhurst]

Great project and implementation.

Have you done any shots you can publish?

What snags did you run into during the build?

One thing I find is that perfect perpendicular alignment is very necessary to avoid a lot of work during the stitch process. I take pains to be within a pixel or so of perpendicularity. I know this is a "technique" thing, but are you finding the same issue? My implemenation with idler was specifically tailored to ease in this perpendicularity alignment. How are you ensuring perpendicularity in this system, or are you using a stitching tool that doesn't care so much?

I've found alignment of camera axis vs rail axes is just as important, but much easier to achieve. It's simple to rotate the camera vs the rails, but much tougher to rotate one rail vs the other.

[mawyatt]

Ray,

Yes alignment is important, so far I haven't done anything special. Just the usual visual, check and correct....certainly not to the pixel level yet! This is not my precision setup though, that's the Thor Labs based system which I'll show soon.

I'll use a 6" silicon wafer as an alignment tool since these are very planar and orthogonal features. I've used them to align the base plate with the vertical bar in the setup shown. Placed 4 very stiff springs between the stand base and the base plate centered around the thru 1/4-20 mounting bolts. By adjusting the 4 1/4-20 Allen Head bolts you can level the base plate with respect to the optical Z axis. I used your PN105 since it's flat to the corners as the alignment measuring tool with the 6" wafer as the alignment sight. This works well, but the 1/4-20 is a little too coarse a thread for this level of precision, these were already part of this plate from Wemacro and thus I just went with this thread. Later I'll probable tap for a much finer thread, or maybe use an insert...but this is way down on my "to do" list!!

The ARCA plate is fixed to the base plate with 4 M5 bolts and has slight room for alignment, as does the base plate to stand base. The THK KR15 have a small amount of "movement" with the M3 mounting bolts, which should be enough to align with. I think I'll be able to get the alignment within acceptable levels with this setup, and the Thor Labs setup should allow pixel level alignment.

I use Zerene for stacking and PS initially for stitching. I found PS to be very difficult to work with for large precision stitches. Rik pointed to PTGui which proved to be much better for stitching. It worked immediately on files that choked PS, and a couple I couldn't get PS to digest no matter what I tried.

I didn't really run into any serious snags building this setup. I spent lots of time thinking about how I wanted things to go together and work, rather than just "jump in" which I often do

The tools I used were just a cheap Harbor Freight drill press , drills and taps. The ARCA plates and clamps were just cheap eBay items as well, so nothing exotic or expensive. The Wemacro Vertical Stand is a great stand to work with and allows easy modifications, this made the construction very straight forward and relatively easy to do.....and it didn't cost much either!! The Misumi 40mm square bar was the only "special" item outside the additional base plate which you can get from William at Wemacro. The Misumi bar isn't necessary for the vertical configuration, but can help with the horizontal configuration if you use a long lens with long WD, but it wasn't expensive either, ~$20 I recall. The entire project, sans focus rails, but including all the electronics, Vertical Stand, adapters, ARCA components, and so forth, probably cost less than $300. Of course I've spent way more to find out what works and what doesn't

Best,

[ray_parkhurst]

I've been using ICE, which is a very simple and intuitive stitcher, but it doesn't like it when the axes are not parallel/perpendicular. PTGui is a bit more forgiving, but seems to be a whole lot more work, and has given me only weird, distorted results. Maybe I'll put some more time into it and see if I can get it to work.

[mawyatt]

Ray,

I only have/use Macs, or now Raspberries , and think ICE is a Windows only program. I haven't given it a try.

Best,

[rjlittlefield]

For ease of stitching, the key elements of alignment are smoothness and reproducibility, not orthogonality.

Any time you have to turn on computational alignment, you create a huge opportunity for adjacent stacks to end up with mismatched Z-axes, maybe even one concave left and the next concave right. If that happens, then for sure you're also vulnerable to stitching defects.

On the other hand, if the Z-axis happens to be just off perpendicular by a degree or two, nothing bad happens to the stitching. The only issue then is that in the final image you get a slightly distorted view of the subject's geometry. That distortion is quite small and almost certainly would never be noticed. It happens to be exactly what you'd get from having a tilt/shift lens be slightly tilted or shifted, or what Zerene Stacker will do synthetically if you put a checkmark on "Generate stereo..." and specify non-zero X- or Y-shift.

So, the keys for trouble-free stacking are (1) to have optics and a focus-stepping mechanism that is smooth enough to allow turning off computational alignment and still get good stacking, and (2) to have the setup rigid enough that the axis alignment does not change from one stack to the next.

In this (excellent) setup, my top concern would be to make sure that the focusing rail is very smooth and is off-axis loaded enough to guarantee that the image shifts no more than a small fraction of one pixel from one exposure to the next.

--Rik

[ray_parkhurst]

For ease of stitching, the key elements of alignment are smoothness and reproducibility, not orthogonality.

Any time you have to turn on computational alignment, you create a huge opportunity for adjacent stacks to end up with mismatched Z-axes, maybe even one concave left and the next concave right. If that happens, then for sure you're also vulnerable to stitching defects.

On the other hand, if the Z-axis happens to be just off perpendicular by a degree or two, nothing bad happens to the stitching. The only issue then is that in the final image you get a slightly distorted view of the subject's geometry. That distortion is quite small and almost certainly would never be noticed. It happens to be exactly what you'd get from having a tilt/shift lens be slightly tilted or shifted, or what Zerene Stacker will do synthetically if you put a checkmark on "Generate stereo..." and specify non-zero X- or Y-shift.

This may be true for some stitching algos (PTGui?), but for ICE it seems to be supremely important to ensure orthogonality. Any offsets show up as stitching errors in the sub-image alignments, and are very noticeable.

So, the keys for trouble-free stacking are (1) to have optics and a focus-stepping mechanism that is smooth enough to allow turning off computational alignment and still get good stacking, and (2) to have the setup rigid enough that the axis alignment does not change from one stack to the next.

In this (excellent) setup, my top concern would be to make sure that the focusing rail is very smooth and is off-axis loaded enough to guarantee that the image shifts no more than a small fraction of one pixel from one exposure to the next.

--Rik

Excellent advice. I have been allowing alignment during stacks...had not considered turning it off, but will do that to see if I can improve. I certainly get a better image turning off scaling, but had not considered turning off alignment. For some silly reason I have always thought that the alignments were to the nearest pixel, and that small alignment shifts would not affect the image quality.

Edited to add: This may be (another) reason to use voice coils for stacking.

[mawyatt]

In this (excellent) setup, my top concern would be to make sure that the focusing rail is very smooth and is off-axis loaded enough to guarantee that the image shifts no more than a small fraction of one pixel from one exposure to the next.

--Rik

Rik,

The Pololu Tic-500 controllers produce very smooth, yet allow quick motor performance if setup properly. I didn't want to have a slow setup since I'll be taking many images, but didn't want to give up any precision either. I played around with a bunch of different controllers and drivers, and these Tic-500 performed the best.

With the THK KR20 I'm using for the Z axis I've found no visual evidence of rail backlash nor wobble, and have not implemented any off-axis loading. I haven't seen any of these effects in any of the THK rails I have.

Best,

[rjlittlefield]

This may be true for some stitching algos (PTGui?), but for ICE it seems to be supremely important to ensure orthogonality.

Speaking as a fellow who has written stitching software, I am puzzled by that assessment. ICE is designed to handle hand-held panoramas and panorama stitching from panned videos. Even their basic Features page shows examples where the tiles are almost randomly positioned.

Lacking any further information, I'm inclined to suspect that you saw some problems that were actually caused by other issues, and attributed those to lack of orthogonality.

--Rik

[ray_parkhurst]

This may be true for some stitching algos (PTGui?), but for ICE it seems to be supremely important to ensure orthogonality.

Speaking as a fellow who has written stitching software, I am puzzled by that assessment. ICE is designed to handle hand-held panoramas and panorama stitching from panned videos. Even their basic Features page shows examples where the tiles are almost randomly positioned.

Lacking any further information, I'm inclined to suspect that you saw some problems that were actually caused by other issues, and attributed those to lack of orthogonality.

--Rik

I don't think so, mainly because I had to re-do several shots due to stitching errors, and when I made sure I had near-perfect orthogonality, the stitch came out near-perfect. Maybe a coin is a tough stitching subject. I know that it's extremely difficult to identify alignment points for PTGui to work with. The features don't have sharp and well-defined points to choose for alignment. ICE may not be able to do the kind of random alignment that it is famous for with coins. Interestingly, in order for ICE to stitch the coin panorama, I must manually adjust the overlap % to within 0.1% or it fails miserably. It is incapable of doing an "auto" overlap.