Historical Airphoto Processing (HAP) with PCI Geomatics

Ingesting the imagery is the first step in your project. We will use the HAP module within Geomatica to create the necessary information for processing, including exterior orientation, flight line information, and a basic ortho engine project we can use to collect fiducial marks off the scanned airphotos.

In order to position the images, we need to build a camera model for each image. By providing the HAP system with the corner fiducial locations (through the collection of a template), the HAP system can assemble a project and begin the alignment process. 

This quality assurance step will roughly position the images to within a few hundred meters of their actual location, depending on the accuracy of the air photo scene center positions provided. Loading the results of the nominal georeferencing allows us to inpect the images and identify any rotation issues, scale issues, or to identify anything that is abnormal.

Having completed the inspection of the images through the nominal georeferencing step, we proceed to the coarse image alignment. We will make use of the reference image to collect ground control points automatically. Automatic tie points will also be collected which we can inspect in the next section.

The coarse alignment inspection includes ensuring there are no images without GCPs (islands). We can use the tools in OrthoEngine to identify problem areas (sorting by # of GCPs per image) and address issues as needed. The inspection step's goal is to reduce the overall GCP RMS to less than 20 image pixels, and to prepare it for fine alignment. You should also ensure a good distribution of GCPs across images and in different types of terrain.

In this step, the coarse alignment is improved upon, and the overall project exterior orientation is edited automatically and improved as well - for example using the block adjusted points, flying height is fixed to ensure the model is as accurate as possible. Similar editing steps are repeated in OrthoEngine to bring the pixel error below 10, or even better (below 6 pixels). Instead of deleting too many points, we choose to pass through the data one more time using another fine alignment process in the next module.

The last fine alignment pass is applied for this dataset, and a final grooming of GCPs and TPs is performed. This ensures optimal results and that the quality of the final model is high enough to attain good quality ortho image production. A good model can also be used to produce elevation models - we choose in this course to leave the pixel error at less than 3 pixels and generate the elevation data. For better models, it is suggested to bring the accuracy down closer to 1 pixel.

Givent that the accuracy of the model was good, we chose to generate elevation data. We created epipolar pairs, then extracted elevation data automatically. The result is a Digital Surface model from 1967, which contains valuable information on the height of forest stands, and structures for example. A demonstration is provided on the use of the Digital Elevation Model (DEM) Editing tool to fix the elevation and also to interactively create a Digital Terrain Model (DTM) which can then be used for the ortho production step is provided.

Having completed all prior steps, we arrive at the final and most important process, to create ortho images, and mosaic them into a single seamless image. We provide you with Python scripts that will allow you to interactively create the mosaic preview, where you can make interactive edits using the Mosaic Tool. Small adjustments are made to the mosaic, such as balancing the images and making easy, interactive edits to cutlines and dodging points. Once the preview looks as good as possible, we export the final product to a new mosaic file. The final step is an inspection of the quality of the corrections, by comparing the historical image with the road center line (provided in the data pack) and also the reference image.