To do it seamlessly the camera would need to change the angle of view as the observer moved towards or away from the screen. It would also need to move from side to side or up and down as the observer moves from side to side and up or down.
In the video they got around most of those issues by placing the screen inside a fairly small space that restricted the possible positions of the viewer. They didn't correct for minor movements of the observer, either.
If you watch the video closely you will see that one of the cameras recording some shots in the video we see (as opposed to the camera feeding the image on the screen) is placed in the near ideal position to match the actual scene to the video scene, but those observing the screen wouldn't be seeing the screen and outside scene from that same viewpoint! You can also see that when someone is very close to the exterior end of the bus stop it totally breaks down as the man talking to his friend on his phone while standing directly in front of the external camera looks like there is nothing between his mid-torso on the screen and his lower legs just beneath the screen.