Canberra, Through Time and Space!
I had a thought after working on some photogrammetry models recently. Could 3D models be made from historical aerial imagery and be used to recreate past landscapes? I have seen other archaeologists lay flat imagery over digital elevation models to get an idea of the past landscape in 3D. However, what if you could generate a point cloud of the area and automatically extract features, such as building footprints or the location of historical watercourses? Gaining this extra information could turn these old purely visual datasets into GIS analysis-ready tools rather than just a visualisation. And it turns out it works!
Canberra 1958, looking to the north-east from Black Mountain
Creating these models required a decent amount of overlapping images over the same area in different periods. After a quick look through the data on Geoscience Australia’s Historical Aerial Photography database (link: https://imagery.aerialphotography.fsdf.org.au/), I found Canberra has been a consistent imaging location since 1945. As such, it would make an excellent location to create a few contrasting models.
Below are a few of the models created for the years 1945, 1958, 1977 and 1990.
INSTRUCTIONS: Click the play button and use the left mouse button to orbit and the right mouse button to move around the model. The models can also be played in fullscreen if that is your preference.
Canberra in 1958
Canberra in 1945
Canberra in 1990
Canberra in 1977
Important considerations in 3d
It is essential to talk about the importance of correctly georeferencing models. Without any control points to scale, rotate and place the model in space, a model is just a visualisation tool. Once you have exact control points put in place by a surveyor, a model can be used to accurately measure distances, draw site plans, place it accurately in virtual space and can be used as a design basis by architects and engineers. Without this crucial step, 3D models are not much more than a visualisation tool. Once they are correctly scaled and rotated, they become powerful investigation tools.
By georeferencing a model, it becomes more than just a pretty visualisation tool to show clients and the public. It becomes a dataset that can form the basis of an investigation or even a digital twin.
Limitations
This data has some limitations, as it is generally the densest over urban areas and pretty sparse over farmland. Consistent models (say, one every decade) are impossible in rural and regional areas. Models can be made, but they are less likely to show the gradual change possible in urban areas. As changes in these areas are far slower, this should not be a huge issue.
The data is also only good when looking at the area in a macro sense. Any kind of investigation at street level, such as walking the streets in a VR headset is going to look pretty bad as the pixels in the original images are quite large. So no historical street view, unfortunately, which would be amazing to have.
Next weeks Blog
In next week’s blog, I plan to review some techniques that can generate data from these models. I mainly aim to analyse how the building footprint extraction performs using point clouds vs deep learning (AI) extraction methods on flat imagery, along with seeing if any other tools within GIS can enhance these exciting datasets.
