element of imagination and creation

Introduction to very close-range photogrammetry

I

Working on my abstract proposition for the CIPA 2019 conference, which involves very close-range photogrammetry, I thought I should write a detailed post about a cheaper way to achieve 3D digitization of small details using this method.

Later edit: working on this post got really really long and I decided to split it up in two parts: one part where I discuss in detail aspects regarding macro photography in another post, and this post here where I focus on the photogrammetry application for macro photography. So, let’s hit it!

What is and when should we use very close-range photogrammetry?

First, let’s see what’s up with this “close-range” and then we’ll get to the “very” part.

You might wonder how many types of photogrammetry are there. Isn’t just “photogrammetry” enough to describe this method regardless of its application? Not quite.

Since its inception, photogrammetry has been defined and categorized in many ways. Gaining adepts in more and more fields and being constantly improved with technological advances, photogrammetry gathered a plethora of labels. Luhmann et al (2006) realized a thorough classification of photogrammetry based on different criteria. In this case, I will focus on his classification depending on the object distance and camera position:

  • satellite photogrammetry – height > 200 km
  • Aerial photogrammetry – height > 300m
  • terrestrial photogrammetry – fixed terrestrial location
  • close range photogrammetry – height < 300 m
  • macro photogrammetry – image scale > 1 (microscope)

Macro photogrammetry is a more likable term, in my opinion, but very close-range photogrammetry is by far the most used term in the literature.

So, close-range photogrammetry refers to all photogrammetry that is not aerial/satellite. It includes buildings, working sites, rooms, objects etc. Note that close-range photogrammetry can also be achieved with UAVs for building facades, working sites again, rooftops etc.

Very close-range photogrammetry comes in when the surface or the details we want to digitize are more or less the size of the camera sensor.

But with a high-resolution sensor we can record details of tens of microns! Without a macro lens.

That’s true. But for photogrammetry you need to fill the frame as much as you can with the object in order to have decent results. So, for objects or areas that are 1-3 cm long you need to get very very close. And to do that you will need special lenses, lenses that have 1:1 reproduction factor. Macro lenses. A 1:1 reproduction factor (or magnification 1x) means that the sensor captures the real life size of the subject.

For your knowledge, in a 1:1 reproduction factor situation, a full frame sensor we can be filled with a ~36x24mm surface while an ASPC-C sensor can be filled with ~22x15mm (Canon) or ~ 23.5×24.7mm (other cameras). In this regard, crop sensors might seem better for macro works but things are not that simple in choosing a camera for macro photography. But this is a discussion for later.

Having set what is close-range and very-close range photogrammetry and how to achieve them, let’s see why and when should we use them.

I’d like to use E.H. Thompson’s (1962) criteria for the use of photogrammetry (in general).

  1. when the object to be measured is inaccessible or difficult to access
  2. when the object is not rigid and its instantaneous dimenson are required
  3. when it is not certain that the measurements are required at all
  4. when it is not certain, at the time of measurement, what measures are required
  5. when the object is very small

To these, Luhmann added three more:

  1. when the use of direct measurement would influence the mreasured object or would disturb a procedure going on around the object
  2. when real-time results are required
  3. when the simultaneous recording and the measurement of a very large number of points is required.

These are conditions under which photogrammetry should be considered, or recommended.

Again, very close-range photogrammetry is to be applied on small objects, or details such as archaeological finds, coins, artwork details etc. Basically anything that is less than 3-5 cm or that requires details less than 0.1 mm (this value is just a personal deduction).

But why won’t you use a scanner?

This is a legit question. In a future post, I will explain more about laser and structured light scanners, the technologies that are now a direct competitor to the digital photogrammetry. Actually, photogrammetry re-emerged from the depths of time as a competitor for these well-established technologies on the market. I would say that photogrammetry is now the under-dog.

Let’s just assume that we don’t have the money to invest in a 25.000$ professional handheld structured light scanner dedicated for small objects, with a resolution up to 0.05 mm.

Let’s assume further that you only have 10% of this to finance your equipment. That’s 2500$. This would be enough for a good brand new APS-C camera (~800$), a vintage wide lens (~100$), adapter and extension rings (50$). With the rest of 1500$, you could get a decent computer to process your data. This is totally debatable. Put another 1000$ and you can build a really good computer there.

But that’s exactly what this is about. You can always improve and change items in your system. Want a better lens, you buy it, you can use it. Want to go macro, get a macro lens. Want to do photogrammetry on buildings, buy a non-distortion wide lens and you can start doing it. Faster, more reliable camera? Get a full-frame. And so on.

I got carried away. This topic can be endless. But fun :).

So where was I? Ah!

Did I peak your interest, now?

Err … I guess? Let’s see an example or something!

Gotta admit I haven’t worked on small objects so far, but I am currently working on something really cool in this regard, but I will present it later. Most likely it will be a tutorial, so keep close!

Still, I’m going to show you my first experiment with very close-range photogrammetry.

Three years ago I wanted to see if macro photography was valid for photogrammetry. Without any prior documentation, I just wanted to apply photogrammetry rules using macro setups on small objects.

The object of choice was a Greek modern coin. Nothing fancy, nothing of historical value. Just a coin.

The setup

In this experiment I used two different setups: an ASP-C sensor camera with a reversed lens and a full frame camera with a dedicated 1:1 macro lens.

Reversed lens setup

For this setup, I used a Canon 600D with an Asahi Takumar 28mm lens (wide lens, AND vintage). The lens was mounted reversed directly on the camera body, without any extensions. For a controlled manual focu, I used a two axes focus slider as seen in the setup below.

Reverse lens mount setup

The whole setup was designed for a vertical shooting, making the manual focus slightly difficult. The magnification resulted was 2.02X.

Tip: One easy way to determine magnification is to take a photo of a ruler at the minimum focus distance and then divide the width of the sensor to the number of mm that span over the width of your image.

Macro lens setup

I don’t have any images to show you but my setup was … handheld. Yeah :).

Handheld, autofocus, natural light … quite the professional.

I used a Nikon D810 with a Nikon 60mm f/2.8g ed af-s micro Nikkor.

Well it was merely a test.

Shooting method

There are two ways to shoot your image set: panning in parallel lines across a planar surface or rotate the object while the camera is fixed at different incident angles.

I chose the panning strategy: with the X-Y slider and handheld respectively. In the images below you can see the calculated camera positions that shows actually my “precise” movements.

With the reversed lens setup I have recorded 162 images of the coin. The distance from the lens to the coin surface was 5.6 cm which is a very good distance because it allowed enough space for surface lighting.

Image overlap for the reversed lens (super macro) setup test

The image above is self-explanatory. The color legend tells us the color codes for how many images are overlapped across the surface of the coin. As seen in the camera positions image, there is an entire row of images “missing”. That’s where I went too much with the row position of the camera.

The areas with light green and orange indicates that there will be a poor reconstruction. This survey report is a good verifying tool in case there is a chance for you to go back and shoot more photos.

With the macro lens setup were recorded … 32 images. The working distance was around the minimum focusing distance (roughly 5 cm from the tip of the lens to the surface of the coin).


Image overlap for the macro lens setup test

The overlaps here were good enough.

Focus stacking was not required because the level of detail of the coin was within the depth of field size even for the 2:1 setup.

The images were directly processed in Agisoft Photoscan (the version of that time …. 2.4 something, I guess). No preprocessing, no masking, I just threw them into the mixing pot and waited for results. Again, quite professional. I know.

The reconstructed models had 132 million polygons for the reversed lens setup and 8 million polygons for the macro lens setup.

For this presentation, the models were both decimated to 1.000.000 polygons.

Reversed lens setup result (2:1 magnification). Digital elevation model. Notice the poorly reconstructed area where there was low overlapping.

Macro lens setup result (1:1). Digital elevation model. Notice the flattening of the details in the central area.

Experiment conclusion

This experiment was not designed as a comparison, but just for the test of the two setups capabilities. There are several aspects that should have matched in order to have a relevant comparison:

  • 18 MPx crop sensor vs 36 MPx full frame sensor
  • 1:1 magnification vs 2:1 magnification

But even so, we can still draw some relevant conclusions.

  • The higher the magnification, the greater the number of images required to cover the same surface are.
  • the higher the magnification and the number of images, the smaller the details that can be achieved => greater model resolution.

In the table below I organized the “numbers” of this experiment. I tried to emphasize the advantages of a low cost setup versus an expensive macro lens.

Compared parameters Reverse mount lens 28 mm Macro lens 60mm
Reproduction ratio 2:1 1:1
Frame life-size width 11mm 36mm
Costs Canon 600D | 50$ vintage lens | 5$ adapter | 20$ slider | 600$ lens
Image set 164 images 31 images
Focus Manual focus (rail slider) Autofocus
Processing time 3h54m 1h52m
3D mesh 118.000.000 polygons 7.566.000 polygons
DEM 4,44 μm/pix 9,81 μm/pix
Lighting and exposure (with different camera sensors) Diffuse white fl. light (2x 5 bulbs) ISO 400 1/30 s f/11 Natural light ISO 2500 1/640 s f/11

To conclude let’s see the PRO’s and CON’s of both methods.

Reverse lens PROs:

  • cheap. really-really cheap.
  • can go beyond the conventional 1:1

Reverse lens CONs:

  • manual focus (not the focus ring, but actually moving the camera back and forth)
  • camera calibration should be resolved manually

Macro lens PROs:

  • autofocus
  • camera calibration autodetected
  • can go beyond 1:1 with extension rings/tubes
  • easier to use

Macro lens CONs:

  • very expensive
  • severe working distance reduction with extension tubes

As a final word, very close-range photogrammetry is a great way to reconstruct in 3D small objects. It is much cheaper considering the alternatives. Even with a 5X macro lens you’d get way cheaper. Versatility would be here the main advantage to laser or structured light scanning.

The disadvantages here would be the manual labor involved, sensitive scaling, pretentious working conditions (lighting and vibrations), focus stacking (tedious process to capture all of the objects’ details in focus).


So this would be it for now.

Soon I’m going to write a detailed tutorial about very close-range photogrammetry with focus stacking and and color correction and everything. Stay tuned!

I hope you enjoyed, see you next time!

Pax!

References and acknowledgments

  • All the images were created by me
  • Meme generated with imgflip
  • Literature references
    1. Thompson, E. H. (1962) Photogrammetry. The Royal Engineers Journal, 76(4) 432–444. Reprinted in Photogrammetry and surveying, a selection of papers by E. H. Thompson, 1910–1976. Photogrammetric Society, London, 1977
    2. Thomas Luhmann, Stuart Robson, Stephen Kyle, Jan Boehm (2014), Close-range photogrammetry and 3D imaging (2nd edition), Publisher: de Gruyter, ISBN: 978-3110302691

About the author

Laurentiu-Marian Angheluta

Add comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

element of imagination and creation

Your sidebar area is currently empty. Hurry up and add some widgets.