Frédo Durand and Soonmin Bae at the Massachusetts Institute of Technology in Boston, with Aseem Agarwala of Adobe Systems in San Jose, California, turned to a technique called visual homing to come up with an answer (ACM Transactions On Graphics, DOI: 10.1145/1805964.1805968). Visual homing is used in robotics to send a machine to a precise location, such as a charging station.
The team's software runs on a laptop linked to a digital camera. The software compares the camera's view to a preloaded historical scene and provides instructions to adjust the camera's position and zoom to best match the scene.
The laptop is a temporary measure, however: "We envision the tool running directly on the camera," the team says.
Perfectly matching snapshots-in-progress with a photo taken in the same spot a hundred years ago is an awesome idea. Turns out, it's kind of hard. But Adobe and MIT have figured out a way to make it happen more accurately.