Working with non-EXIF photos

[jerry7171]

Hello Everyone,

I'm extremely new to the field of photogrammetry. I'm reading everything I can find online and at the library at the moment. I think this will be an exciting new side hobby for me along with my love of photography and history.

I've been playing around with a trial version of Photoscan and getting my head wrapped around a world of new concepts and terms. I am confident that I'm starting to catch on but I have a big question not unlike some of the others I found here:

Is it possible to take non-EXIF photos from the past 150+ years and use them to model objects or scenes that no longer exist or have changed substantially since they were first photographed?

As an example, I'm combing through one of my favorite free archives, the Library of Congress, and was inspired by the cliff shrines and temples at Silsila in Egypt: http://www.loc.gov/pictures/item/2004666763/



There are a number of other photos from around the same time frame elsewhere, but obviously the big obstacles to overcome will be the fact that all the photos were taken at different times of the day from odd locations, are of varying sizes and resolutions and black and white.

I can immediately read each image and discern a number of features from photo to photo that would aid in creating a model.

I'm under no illusions that Photoscan can recreate Silsila as it then existed down to the last pebble, but an overall crude model would be fun to start with.

Am I jumping in over my head by attempting this sort of thing right off the bat with little experience or is the software going to have trouble working with these sorts of images? Is there anything I should do prior to attempting my recreation of the temples and topography?

As an aside, I've noticed that there are photos available at Flickr that are free to use for private purposes under the Creative Commons and I've thought about bringing a number of those over to pad out the small number of photos, but I hesitate since they're cleaner, sharper and in color with EXIF and I don't think mixing the two types of images would be very helpful to Photoscan.

Happy Holidays everyone, and I look forward to everyone's advice.

[Alexey Pasumansky]

Hello jerry7171,

I think that your task is similar to the one already mentioned recently in another thread:
http://www.agisoft.ru/forum/index.php?topic=1776

There are several important things that should be noted regarding old scanned image processing:
- lack of EXIF data (however, I believe that it is not critical),
- photos could be retouched and in this case some real features could be lost and new features could be induced,
- original images could be cropped before printing and scanning,
- scanning/printing artifacts,
- for some objects/scenes there are too few camera positions old images were taken from (like "the best view" positions), and for PhotoScan it is preferred that images are taken from different positions to provide sufficient stereobase,
- different light/shadows conditions on photos from different times,
- the object/scene itself could have changed during ages between images were taken.

But I think that some results could be reached even with all that mentioned above. Probably, you'll need to use markers that should be placed on the same point across the whole image set, of course, you'll need to use split camera calibration mode. Possibly masking out background area will also help.

[maddin]

There are several important things that should be noted regarding old scanned image processing:
- original images could be cropped before printing and scanning,

Not to mention the use of old view cameras (lens moves independently of the film) or tilt-shift-lenses sometimes use to photograph architecture... I would expect both create big problems in PhotoScan, no?

Martin