Attempting to reconstruct a 2 meter high, 25.4 cm diameter cylinder

[Modeler17]

I just downloaded a trial version of Zephyr light. My first project is to reconstruct a cardboard cement mold cylinder with writing on it and a set of round plastic balls hanging from one side. I put the cylinder (upright) on a turn table and seemed to get nice coverage of it (with recommended overlap) in 28 shots. The turntable does about one rotation every 25 seconds.

I then masked each photo so just the cylinder and hanging balls are visible. Running it with mostly default settings in close range mode, with circular photo ordering selected, seemed to give the best results. However, only half the cylinder gets modeled. I am looking for suggestions as to what I should check.

[Scott Nebeker]

28 shots for something of those dimensions should provide adequate coverage. In fact, it could be too much coverage (judging by the diameter of the cylinder alone). I know it's odd to read this but sometimes less is more when it comes to photogrammetry in general.

First thing I'd suggest, based on what you write, is to verify the source imagery. Going back to the source imagery is always the first thing to check. From what you write, it appears the turntable is in constant motion. Is that correct?

If so, I'd suggest looking for motion blur in your images. It could be throwing Zephyr for a loop because the images are poorly focused due to the effect.

--This whole next part, is long and only partly explains why I'm thinking this is a problem in your case. It's nerdy and potentially impossible to grasp because of my poor ability to explain things You don't need to read it.--

I'll play this out by assuming the turntable is in constant motion. I'll also round your numbers up. Let's put it at one rotation every 30 seconds and 30 shots in the same amount of time (one shot per second). With those numbers, a single rotation will have the cylinder's outer, visible edge traveling 12 degrees per second, assuming the cylinder is dead-smack centered.

I'll borrow Kepler's third law of planetary motion because it best describes how fast this sucker is going. The law says:

[em]"The square of the orbital period of a planet is directly proportional to the cube of the semi-major axis of its orbit."[/em]

Replace "planet" with "thing" and BLAM! we've got math! From there, we can use Kepler's second law to figure the speed. But, we don't need the math, we just need the concept. Imagine Earth and Neptune had the same orbital-period (year). Earth would be cruising along at it's normal leisurely pace around the sun while Neptune would be traveling incredibly fast in order to make it around the sun in the same 365 days. It would be going SO fast that it would lose cohesion and disintegrate into a streak of bluish gas where there once was a spherical planet.

To put it simply, the further away a thing is from the thing it's going around, the faster it must go to cover the same degree of arc.

So, a 10 inch diameter cylinder that's dead-centered on a turntable that's in constant motion at a speed of 30 seconds per revolution. That makes the outer-edge of the cylinder 5-inches from the center, which means (taking all that math into account) it's spinning in front of your lens at.... "pretty quick" is the solution my calculator tells me. "Pretty quick" is okay if your camera can account for it with an equally "pretty quick" shutter speed.

Yeah... that was WAY too nerdy. My bad.

[Roberto]

Hi Modeler,

I agree with Scott, I just want to add another hint that might help.
If the cylinder does not have a lot of texture, you can try to add some 'texture' in the turntable (for example a newspaper) and include the turntable in the masks. This will help the first phase of camera orientation.
After all the cameras has been oriented successfully, you may optionally change the masks (tools-]workspaces-]change maks) to exclude the turntable before running the densification process.

[Modeler17]

Thank you for your suggestions Scott and Roberto. And I even think I understand Scott's solar system analogy.
To answer some of Scott's questions:
The turntable was indeed in constant motion.
Regarding the possibility of motion blur, My photos indicate that I successfully eliminated motion blur with my lighting strategy. Because maintaining a high enough shutter speed at f8 while maintaining a low ISO requires a relatively large amount of light, I used a flash bounced of the ceiling mounted on my DSLr's hot shoe. The flash was sufficiently brighter than the ambient light which effectively stopped motion regardless of the 1/60 sec max shutter speed my camera allows with flash. The other good news was that the flash did seem to keep up with the 1 exposure per second shooting speed I was using. Bouncing light off the ceiling made for even and consistent lighting across shots.

I will repeat the process using texture on the turntable as Roberto suggests only I might use a human model instead of a cylinder since my goal is to do full body scanning. Along those lines, I wonder if there is a way to see the parameters values used in the human scan preset? I have more questions about body scans but I think they might deserve a new topic.

[Scott Nebeker]

Oh my, oh my, oh my!!! You understand photography at a much deeper level than I was expecting. This is very helpful.

There was also the issue of you using a camera with a "rolling shutter" that I wasn't going to get into. Having you know what a DSLR camera is and how to use it means we can talk the same language and I can give a new set of suggestions.

In the user manual for Zephyr, Roberto has outlined certain aspects of the sensor-size and importance of pixel-size. Most DSLR cameras will kick out images at a high relatively high megapixel. There are times where this is useful and times when it's simply not. In this case, you might be running into the issue where less can be more. 28 shots, like I mentioned, is adequate but if you're using the full-size images that are straight out of the camera, it could be too much. There are times when I'll take the images into Lightroom and reduce them by 50% or so (depending on the reconstruction). Doing this, allows for a larger pixel and a smaller file-size. This gives Zephyr, for a lack of a better term, a larger target to hit with larger pixels and shortens the render-time needed for the algorithyms to do their work. Larger targets are easier to hit, shorter render-times are easier on YOU because you can whip through iterations of the modeling process to fine tune it.

As for the way of seeing the parameters for the presets, simply select the preset, then change it to the "Advanced" or "Custom" options from the drop-down. From there, you'll see what the parameters are. Essentially, changing the Preset will change the Advanced and Custom options accordingly. This is what I use in order to fine-tune the Samantha stage to my needs/liking. It's how I've seen the software work and Roberto may well call me out on that and correct me.

As a note here, I've tested almost every piece of photogrammetry software before deciding on Zephyr and the flexibility given to the user like this is the primary reason I've chosen it over the others.

EDITED TO ADD: I'm curious if you'll share the dataset. Is that something you're comfortable with?

[Modeler17]

Hi Scott Thanks so much for following up.

Yes, I can certainly share the dataset. I just need directions as to how best to do that given 28 images at about 1.7mb each and 28 mask files.

As I usually do, I shot raw and used lightroom process photos. Unsure exactly how to set the LR export parameters for this project, set them to include exif (of course) and export as a highish quality (7) jpg. Does this sound right?

What actually happened in the course of Zephyr camera placement seems to be that the CONVEX back side (not visible from actual camera position) of the tube was mistaken for a CONCAVE surface kind of plastered to the inside of the front surface. So the tube ended being a halve tube with an inside surface lined with the outside surface texture. Also the two hanging balls were ghost like and fractured, scattered around the surface of the image. So I'm wondering if the repeating pattern of print on the tube confused Zephyr during matching somehow.

I will continue to experiment. I am glad to have your opinion about Zephyr since you have experimented with so many photogrammetry softare products.

[Scott Nebeker]

I've actually used and broken software from other major players in photogrammetry and I've got mixed responses. Two of the companies are ones that produce MANY pieces of software and they opened their entire libraries to me. Both of those companies have responded negatively to my public feedback even though they knew I was pitting their stuff against others in a public way. It was a good little excercise and showed me a lot that I didn't know.

As for sharing the files, I can open a shared drive for you or you can do the same using your Google account (which comes with a good chunk of storage space in Google Drive).

Right now, I think it's best to wait for Roberto or Andrea to weigh in again to be sure I'm not leading you astray. They're both in Italy and it's currently past 1am for them.

[em]((Please notice that I'm bowing to those two. THEY (and some others) speak officially for 3D Flow, I don't. I'm just an obscure guy from an obscure place in the US. With you using the trial version, I think it's important that you get solid results before your time is up))[/em]

[Modeler17]

Entire libraries,Wow, its great that you have been able to develop such a detailed understanding of the way photogrammetry software works.
OK, I will wait to see what Roberto or Andrea say regarding the best way to share my image and mask files.

[Scott Nebeker]

Oh... Please don't take what I say about companies giving me full access to their libraries as a sign that I'm perfect at this. I've noted that some of this is trial and error. It's true with every piece of software out there. Just because I've tried a lot of solutions doesn't mean that you shouldn't. Every person has their own preferences, I've found that Zephyr meets mine. That doesn't mean that Zephyr will meet yours.

Most of the companies that make photogrammetry software will give you a trial version. I would say that you should also give those a try. My opinion on the matter is just that. Nothing more.

[Roberto]

I'm going to create an account on our server so that you can easily share the data with us. I'll send you the login information by email.

[Modeler17]

OK Roberto and Scott, I have uploaded my data. Please let me know if there is anything else I need to do in order to allow you to see the files.

[Modeler17]

So, Scott, it sounds like I might have more body scanning success going for 3 towers with 3 cameras each instead of 4 towers with 2 cameras each?

[Scott Nebeker]

[Modeler17]

I apologize for not quite knowing how to share my files.
Here is a link I generated.
Cylinder Project
If this doesn't get you access, I might need explicit instructions on how to allow a forum member access.

One thing I don't quite understand in your last post is the concept of stopping the turn table at intervals in order simulate towers of cameras. Would you be willing to explain how this would work in more detail? I am in the process of deciding how many cameras I will need (I don't have 8 cameras yet, but I do have good turntable(s)) and costing out the project. Understanding this could reduce the expense of the project significantly, and therefore make implementation more likely.