the viewer appears to be directly above every point of the image simultaneously. It is an orthographic perspective projection, viewed from infinity. There is no foreshortening, and the scale is constant throughout the image. Objects closer to the camera don't look any bigger. The only way something can be "behind" a mountain or obscured by relief is if that spot on the ground was visible in only one of the stereo pairs, but not on the other.

In aerial photogrammetry, the two photographs in a pair overlap by 60%, and the area in the overlap zone is visible in 3D. However, this camera geometry exaggerates the vertical dimension substantially. This makes it ideal for elevation measurements and contour mapping, but the visual appearance of the ground is not very realistic. Trees and buildings appear to pop out at you, much more so than if you were in an aircraft looking out the window.

This used to be done by the plane flying in a straight line over the area to be mapped, the camera exposures timed so that the individual negatives overlapped by the required 60%. Repeated flight lines "sidelapped" so that every spot in the mapped terrain appeared in at least two, and sometimes three frames.

Nowadays, I suppose they use digital line scanners and DTEDs (Digital TErrain Datasets) acquired by LIDAR to provide the Z-component. It used to take two years of training to be able to see in stereo well enough to do the work without supervision. Like everything else I once was really good at, the skills required to do the job have now become obsolete.

Getting old sucks.