Google Street View, while very useful, fascinating, and full of wonderful bloopers, does rub some privacy advocates the wrong way. Should people on public streets have a reasonable expectation of not ending up with their photo on the Internet? There’s a whole other article in that, but in any case... for all the folks who do have a problem with it, a computer science graduate student is working on a solution: software that digitally removes pedestrians from Street View images. One of the byproducts of the current version of the system is somewhat unsettling, however – areas where people were in images are sometimes marked by ghost-like shapes, or even by disembodied shoes and feet.

The as-yet-unnamed program was designed by Arturo Flores from the University of California, San Diego, as a proof-of-concept for CSE 190A, a project-based computer vision and machine learning class. While Street View does already blur faces, Flores feels that clothes, body shape, and height combined with geographical location can still be enough to make some pedestrians identifiable.

The system works by identifying human forms in each shot, erasing them, then filling the space in with background imagery obtained from the shots taken immediately before or after. It can be stymied, however, in cases where someone was walking in the same direction as the Street View camera car, at such a speed that they blocked the same bit of background for several shots in a row.

The system also only works in urban settings, where backgrounds are predominantly flat.

Flores used the Street View pedestrian locator, designed by computer science professor Bastian Leibe from Aachen University, as a jumping-off point when creating his system. He now plans on improving his pedestrian remover so it can identify and remove not only individual pedestrians, but whole groups of people.