Grad Student Builds Google Street View Pedestrian Remover
Flores and USCD professor Serge Belongie outlined their system in a paper entitled 'Removing pedestrians from Google Street View images' (PDF), which they presented at the IEEE International Workshop on Mobile Vision held in June. According to Physorg, the still-unnamed program erases human figures from Street View shots, and replaces them with a pixelated approximation of the picture's non-human background. On occasion, however, the system can leave behind some weird, if not chilling, photographic residue. In Flores's world, for example, humans may be an extinct species, but their leashed dogs or shoe-clad feet are not.
Flores and Belongie also struggled to wipe out figures moving at a certain speed and in the same direction as Google's camera-equipped car. Under these circumstances, a single pedestrian may block out the same spot of background in more than one frame, thus making it substantially more difficult for the program to swap pixels. Moreover, the system only works on photographs taken in urban settings, where background pixels are usually "on a dominant planar surface," and easier to replicate. Although the program could easily erase a person standing in front of a mural of a countryside setting, it wouldn't be able to block out a pedestrian standing in front of an actual rural background, simply because the setting is not predominantly flat.
The system, then, may not be perfect, but it probably doesn't have to be, either. Google's primary concern, after all, should be placating its critics, and this seems like the kind of mechanism that could help do just that. Privacy advocates may be a finicky breed, but it's hard to imagine anyone finding fault with a system that, at worst, reveals only your shoes to the world. [From: Physorg]