Orthorectification of historic aerial photos

Updated Dec 14 2020

One of the occasional tasks I undertake in my geological mapping work is to orthorectify historic vertical aerial photos in order to examine them in a GIS. The old photos have much value as they capture the landscape at particular point in time where it may have been subsequently modified. I have recently been working in the Launceston area and have created mosaics of the oldest and most detailed photos available in order to show extent of older quarries, geomorphic features, geology, etc. As an example, I was recently asked to comment on a development application on a former quarry site and the old photos were very good at showing the extent of ground disturbance. I created a digital stereopair with StereoPhotomaker (see earlier blog) from the scans and made a rectified image for the GIS from which I was able to gain a good understanding of the site.  

This blog provides a documented methodology for my own benefit and anyone else.

There are two main considerations based on extent. If I am interested in a small area (i.e. within the extent of a single photo) I will simply use a single scanned photo and employ the Georeferencer plugin in QGIS (or ArcGIS equivalent) to stretch the image over a modern orthophoto. This is usually good enough in areas of low relief where accuracies in the order of 1-2m can be achieved. However, if I am interested in creating an image of a large area covering several photos then a different approach is required. My go-to tool is Agisoft Metashape (formerly known as Photoscan) that is widely used globally and by the University of Tasmania in particular. Metashape is propriatery software and comes with a manual and video instructions. In this blog I will outline a few specific steps that are not documented when dealing with historic (film-based) aerial photos. My experience is related to obtaining digital scans of negatives and prints from Land Tasmania, DPIPWE but I expect it will be relevant for other jurisdictions.

1.     Ordering: Obtain a set of photos that extend beyond the area of interest by at least one frame if possible. I try to obtain the largest scale and oldest acquisitions available at 600dpi or better. TIF format is preferable.

2.     Clipping: All of the images need to be exactly the same dimensions (X and Y pixels) for Metashape to work. Unfortunately scanned images for a given acquisition tend to vary slightly and you will see this reflected in the image details in your computer's file manager application. In the file manager, order your photos from smallest to largest. I then use the Irfanview programme (freeware) to examine the pixel dimensions of a few of the smallest images in the photoset. I note the X and Y values for a few photos until I am satisfied I can identify the smallest dimensions for both axes. I then use these values to crop each photo using the batch processing capability of Irfanview with the cropping occurring about the centre of each photo and writing these cropped images into a new folder. This process will trim a little of the frame on the larger images but shouldn’t affect the photo itself. You may also need to consider rotating some images in order to create a consistent orientation with respect to the instrument panel (see masks below).

3.     Loading into Metashape: I will not go into much detail here as the specifics are documented in the help file. However, the general principle though is not to enter camera properties but leave that for Metashape to work out for each image. 

4.     Masks: Masks should be made for each image to remove the instrument panel, fidicials (edge ones only) and the frame from the processing. If all the photos are oriented the same way with respect to the instrument panel, then one mask can generally be applied to all the images in a single operation. Therefore, take care in creating a mask that will work on all your images accepting that you may loose a little of the photographic image close to their borders. The workflow includes exporting the generic mask as a template image then importing it to be applied to all images. Take your time with the import step as it is easy get this wrong and get an error.

5.     Markers: Run the Align Photos function from the Workflow menu (e.g. on a Low setting to make this quick). Once aligned check to see whether the camera positions and the point cloud appear to be in the correct alignment. If not then you may need to place markers for errant photos and their neighbouring images in order to help the software. Rerun the alignment function until satisfied. If you have a lot of large images you may consider dividing the photos into chunks

6.     Exporting Orthomosaics: I continue through the worksflow using the default settings until such point where exporting the orthophoto is possible. Rather than entering real world coordinates in Metashape, I export in local coordinates only as I prefer to do the georeferencing part in the GIS.

7.     Georeferencing: I find that only a few tie points close to the margins (but not on the extreme edges) are required to adequately georeference the image in the GIS programme of your choice.