Jump to content

Planning for High-Resolution and Multispectral RTI Capture and Photogrammetry


Taylor

Recommended Posts

For capturing high-resolution multispectral images of a 19th century landscape painting (36 x 50 inches), I found it helpful to create a spreadsheet for planning purposes, and also to supplement the shooting notes as a record of what was done. To get complete coverage of the painting at the desired resolution for RTIs and photogrammetry (500 ppi for RTIs, and up to 2600 ppi for certain details), it was necessary to take a series of overlapping images.

 

The spreadsheet takes basic information about the object (size, material, UV sensitivity), camera and lens (format, sensor and pixel dimensions, focal length, and settings), and calculates various parameters (working distances for the camera sensor and light source, minimum size of the reflective spheres, number of images, and storage requirements) for the project, given the desired target resolution and various wavebands (UV, visible, and IR) to be captured. This information is useful for estimating the space and time requirements for capturing RTIs and photogrammetry. Since the painting is on the east coast of the U.S. and I'm in California, it was important to have a good understanding of these parameters before shipping equipment across country and for arranging studio space in which to do the work. The spreadsheet was also helpful for selecting the macro lens for the project.

 

The storage requirements are based on a RAW image file size of 20 Mb (a slight overestimate for my 16 Mp camera) and don't take into consideration the processed file sizes. For example, generating .dng files with embedded RAW images approximately doubles the RAW file size, and exporting .jpg images adds approximately 50 percent to the storage. The final processed .ptm and .rti files range from approximately 250 Mb to 350 Mb per RTI, so accordingly, additional storage will be needed to process the files. The spreadsheet only estimates the storage needed for RAW image acquisition.

 

Another variable is the amount of overlap for the images. For general imaging and RTIs at a given resolution, the spreadsheet uses 10 percent overlap, and for photogrammetry, it uses 66 percent overlap for the camera oriented horizontally. The spreadsheet calculates the distance to shift the camera in horizontal and vertical directions to get complete coverage of the object. It assumes three images per position for photogrammetry (horizontal and two vertical orientations) and 36 images per position for RTIs. These parameters can be adjusted for particular project needs.

 

The input parameters are entered into the spreadsheet using metric units. A companion worksheet mirrors the format of the metric spreadsheet and automatically converts all the distances from metric to English units, for convenience. An example of the spreadsheet is attached, showing the calculations for this project. [see update below.]

Link to comment
Share on other sites

I've been using the Arc3D web tool (http://www.arc3d.be/) to process the photogrammetry. The Arc3d tool doesn't use the lens calibration shots in vertical orientation. My understanding is that it relies on greater overlap between images, and generally requires 90 percent overlap to get good results. Due to time limitations, I shot the photogrammetry of the entire surface of the painting with 66 percent overlap at approximately 350 ppi, whereas the RTIs ranged from 500 ppi for the entire surface to about 2,500 ppi for some details. I used the spreadsheet as a guide for planning, but it doesn't capture every detail of what was done, which I try to record in the shooting notes. The spreadsheet tends to slightly overestimate the number of images and storage requirements.

 

I was able to get a photogrammetry model of a partial fingerprint in the painting at a resolution of about 2,500 ppi using Arc3d, using 10 horizontal images with 90 percent overlap, without using any of the vertical calibration images. I captured the vertical calibration images anyway, to have a complete data set.

 

Using Arc3D to model the photogrammetry of the entire surface at 350 ppi has been a little more difficult, possibly because I used 66 percent overlap (90 percent overlap would have required about 700 images, a three-fold increase). For 66 percent overlap, the photogrammetry involved capturing 9 rows of 12 horizontal images, plus 24 vertical calibration images per row.* Arc3d produced a result for these images, but I haven't been able to view them using the low-resolution on-line viewer. It looks like I will need to learn to use Meshlab to process the full-resolution model.

 

I haven't decided whether to purchase software for processing the photogrammetry. I'd be interested to hear what others are using and whether there are additional open-source solutions available. Thanks for your interest!

 

*Although photogrammetry doesn't strictly require vertical calibration images at every overlapping camera position, Mark had suggested that I take these additional images. It didn't add a great deal to the storage requirements at 350 ppi and 66 percent overlap, and the additional data may be useful in the future.

Link to comment
Share on other sites

One of the problems we've found with very close-range photogrammetry is that the focal length of the lens is usually too long to produce good base-to-distance ratios. We routinely use the Nikon 60mm macro for close-in photogrammetry, but the accuracy can suffer when the base-distance ratio is out towards 1:7-1:10. You also get razor-thin depth of field with this sort of work (perhaps not a problem with a painting). With higher resolution cameras you can pull out a bit farther with a wide angle lens, like a 28mm can get roughly the same results in terms of point density but with more accuracy (better base-distance ratios).

 

Do you have any sense how many points Arc3D is generating per inch? I haven't used that tool in quite a while. The free and inexpensive photogrammetry packages, like Arc3D, My3DScanner and 123D Catch seem to all do camera calibration by a sort of brute-force method. Under optimal conditions Agisoft can produce some nice results, although once again it does not allow access to low level calibration. There's a new incarnation of Bundler, called the SFMToolkit, that looks quite powerful: http://www.visual-experiments.com/demos/sfmtoolkit/

 

Thanks for these really interesting posts, Taylor.

Link to comment
Share on other sites

Thanks for the pointers to other sources of photogrammetry software. I was aware of Bundler, but hadn't heard about the new version of their software. I wonder if there will be a version for Macs or 32-bit Windows OS. It's a lot to ask of open-source developers. Unfortunately, I have a 32-bit Windows XP system and a Mac Mini to work with, and neither has an Nvidia graphics card. I hope I don't need to go shopping for another computer!

 

It appears I'll have to learn to use Meshlab to find out the actual resolution of the photogrammetry produced by Arc3D. I haven't yet downloaded the full-resolution data. The resolution I gave above was just for the images, not the 3D models.

 

Regarding the base-distance ratio you mentioned, I've incorporated this ratio into my spreadsheet [updated version attached]. I used a Lumix micro four-thirds camera with a 45 mm Leica/Panasonic lens--the equivalent of a 90 mm lens on a full-frame DSLR. It will be interesting to see how this combination affects the accuracy of the models. My camera was modified by LifePixel (www.lifepixel.com) to allow UV-Visible-IR photography.

 

Thanks for your interesting and helpful comments!

RTI Res Calc Draft 11-28-12 Res Calcs (m).pdf

RTI Res Calc Draft 11-28-12 Res Calcs (in.).pdf

Link to comment
Share on other sites

  • 5 months later...

I've uploaded the Excel version of my spreadsheet, since Carla recently changed the forum to allow uploading of Excel files (.xls only, not .xlsx).  This should be more useful than the .pdf versions I uploaded previously.  The spreadsheet would need to be modified for different cameras, sensors, and lenses, but this should be easy to do.  Just keep in mind that many of the cells contain formulas to calculate distances or to convert from metric to English units.  Of course, the spreadsheet comes without any warranty...etc.  I hope others will find it useful, and let me know if you have suggestions for improvements.

 

Update:  I changed the minimum sphere diameter to 250 pixels from 200 pixels (cells R18-R25), to reflect current recommendations by CHI.  This is because not the entire upper hemisphere records the light positions, due to the optical geometry of the spheres.  A revised spreadsheet is attached.

RTI Res Calc Shared Draft 23-Apr-13.xls

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...