Hi Chris. Daunting task on your plate. Napkin math indicates you have 2500 images / year on average. Perhaps start with a year and create a point file from centerpoint as a kml linked to image. I'm not sure how to create an ESRI relational geodatabase from scratch, but I imagine it's possible. With either the KML (Google Earth format) or ESRI you could have popup-enabled points/feature classes. In the former, basically a Google Earth file with popups. With ESRI you could create Viewer Attachment Instant App.
These are preliminary steps to wrap your head around the data. To georegister those images would require creating tie points to co-register each image. I imagine there are advanced object classification techniques to bulk classify images to enable an automated process, but that would require remote sensing and/or AI skills. Not to dissuade you. Ultimately though for manual-ish rubber sheeting techniques, you need multiple tie points in your images with known real world coordinates. From there you can automate your process.
Not sure about ESRI, but GDAL has a core utility called gdal_translate that can be used to warp your images to ground control points (i.e. tie points). Something like this:
gdal_translate -gcp 507 482 6246215 2268541 -gcp 1594 289 6246103 2268561 -gcp 3000 674 6245981 2268666 test_projected.tif test_projected_gcp_3pts.tif
gdalwarp -order 1 test_projected_gcp_3pts.tif test_projected_gcp_3pts_warped.tif
That's a command line call above, wherein the first line "-gcp 507 482 <easting> <northing>" is translated as the pixel at column 507 and row 482 = the easting northing coordinate. "test_projected_gcp_3pts_warped" is the resultant output and will be "rubber-sheeted" to align your ground control points. There is a protocol to iterate over those two command line calls from a csv containing those coordinate values in bulk. So essentially if you can build a shapefile, csv, etc. with your matching ground control points, you can theoretically georegister / georeference multiple images in bulk. The time consuming part will be generating the ground control points.
Note that GDAL is available via Python in Arc Pro. So via Python Code in a Notebook might be nice.
There are certainly other methods used in high resolution topographical image geo-registration. Perhaps someone at School of Mines is familiar with his python packages . The same concepts could be applicable imagery, but it's tough since drastic change has occurred over those 70 years.
