2

I have a set of 1200 historical aerial photos covering North Carolina outer banks area from 1974, 1984 and 1992. I need to georeference them and produce seamless color balanced dataset that will be utilized for compilation of NC outer banks atlas. Except for scale information, flight number and date on photo paper copies, no other information regarding camera positions or ground control points is available. No matter how well I pick control points for image-to-image rectification (I use ArcGIS base map for that), the photos still don't align properly when I display them together in ArcMap. I learnt that analytical aerial triangulation may significantly improve the results, but it seems like it needs GCPs that I don't possess. So, my question is how can I triangulate my photos (or fix my alignment problem in any other way) without GCPs and preferable using ArcGIS, Agisoft Photoscan or other open source software.

enter image description here

1

In my work I encounter this situation often. The method that I use is next:

  1. Prepare the images in Photoshop: crop, contrast, brightness, whatever is needed to improved image quality.
  2. Also in Photoshop use the Photomerge tool from File->Automate->Photomerge. Keep "Blend images" checked. After testing it seems that the "Auto" method yields the best results when it comes to georeferencing. Here is a short explanation on the merging methods. If you are going to use many images at once you might get and error with "scratch disk full"; just add another partition with a lot of free space, because Photoshop is out of TEMP memory.
  3. Next step is georeferencing; if the place is completely reworked, you'll have a tough time finding common points. If you are lucky and the settlements still exist, then the roof corners, road intersection, any kind of building will help a lot and will make the georeferencing processes easier. Regarding the RMS, I try to keep it under 2 meters overall and the most often used transformation is the 2nd Polynomial. First Polynomial it's just a matter of luck and third Polynomial, when I have a lot of points on which I am confident, but the error is still high and the image does not want to align.

As a rule of thumb I merge images where the area has been almost completely reworked and having to georeference multiple small images is impossible as there are places where nothing is left from 60-70 years ago. Where the landscape remained almost the same I prefer to keep the images simple because on a smaller image, the error will be smaller and if one is careful while doing the georef the overlap will be almost perfect.

Another solution is to use ERDAS IMAGINE and from toolbox use the Autosync Workstation. It's a nice tool, the only drawback is that you have to convert to the .IMG file format for the best results.

Now it all depends what kind of work do you intend to further do on the old imagery. I work on UXO(unexploded bombs from WW2) where precision is important, the lower the RMS, the better.

  • Hi! Thanks a lot for the answer! I was tempted to do it in photoshop first, but I wasn't sure if it's okey to do it this way. Thank you, now I see that it may work much better for my purpose. Precision does matter to me because the idea is that you can swipe between photos taken at different years in the end. Also thanks for ERDAS tip, I will definitely try this tool. – Ann K. Oct 9 '17 at 13:13
  • Also, forgot to mention, ArcGIS Pro has an auto georeferencing tool that works very good. It's applicable in case you georeferenced and image,with many tie points, an urban area for example; add the next overlapping area adjust the image size to the georeferenced one and just place it, more or less on top and hit auto georef and you will get a seamless overlap. – alecsx Oct 11 '17 at 6:47
0

Years ago when I did Photogrammetry you looked for tie points. These were just points that were identified on two or more photos, to build a photo model. We used points that were on sharp features like a roof corner or any sharp edge. Once we had points to hold our model together, we use GCP to tie the model to the earth/map. Basically these were easily identified points that were surveyed or the coordinates were given to us. Usually white X marks for us on the photos.

Now with GPS you can get GCP of any feature, if not pull coordinates off a basemap ortho, understanding the ortho errors will influence your errors. But the georeferencing is the same, find good features on both the base map and the aerial you want to register, You may need many points depending on the scale, camera errors, and elevation. Realize the ortho already has the camera and elevation error adjusted. You need to adjust/warp your images to an ortho to make them look/fit the basemap aerials better. Play with the different type of transformations in ArcMAP and you should end up with a usable product.

  • Hi Bill, thanks a lot for the answer! I see what you are saying, I tried transformations and other tricks, but the closest I can get is 2 meters error. I wish we had surveyed GCPs.. The challenge with the photos is that shapes and features of the outer banks dramatically changed over the last 40-50 years and it's extremely hard to find proper control points for georeferencing. Another thing is that 50 percent of the image is water, and thus I can't distribute control points evenly. What is the acceptable RMSE error in that case? I'm not sure if 2 m is okey, or I should try harder to do better? – Ann K. Oct 2 '17 at 15:12
  • You are not using ortho software which accounts for elevation which may not make much difference in a coastal area. I wouldn't worry about warping in water, If you tie an image to ground features like road/trail intersections and buildings well then you can use features like shoreline from one image to another to tie them together. Realizing shoreline in image to ground may not work because of change. If you are off by 6 feet and the images covers a mile (5280 feet) that's pretty good. Most state orthos have a 1 foot pixel resolution but can be off 5 feet and still be considered acceptable.. – Bill Chappell Oct 2 '17 at 17:19
  • Thank you Bill! That's great to know, I'm getting decent accuracy then.. Still wish there be a way to get it close to 100%. – Ann K. Oct 3 '17 at 15:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.