It all starts with high-resolution detailed imagery captured by specialised cameras installed in winged aircraft and helicopters and on ground by cameras mounted on cars of pre-determined sites/routes.
These images are then scaled and corrected using photogrammetry a process of determining the geometric properties of objects from photographic images.
Using existing software, a dense 3D point cloud is created essentially length, breadth and depth co-ordinates located in the real world. These points help create what's called a 3D wireframe model which acts as a skeleton. Using the actual photographs, the 3D models are then textured or made life-like.