Software Removes People from Google Street View

Pesky pedestrians can be removed from Google Street View with some experimental software developed by a graduate student at the University of California, San Diego.

Google Street View creates its panoramas of places by stitching together images. However, the images are captured indiscriminately from a moving vehicle. If someone's walking their dog or just taking a walk, they can appear in the final panorama.

Currently, Google blurs out the faces of people, as well as license plates on cars, that are accidently captured in Street View, but the proof of concept software demonstrated by Arturo Flores and UCSD computer science professor Serge Belongie at the IEEE International Workshop on Mobile Vision held in June promises to wash human artifacts from the panoramas entirely. (Flores is a former Senior Application Developer for PCWorld and we're darn proud of him.)

What the software does is identify human forms in an image and removes them. Then it fills in the gaps with pixels from whatever's behind the person--walls, grass, pavement, bricks and so forth--taken from frames shot before and after the one the person appears in.

Similar techniques are used in photo editing programs, but they require a certain amount of manual manipulation. Flores' software does a good job of automatically washing people from the images, but if the photos are closely scrutinized traces of humanity can still be seen in them. And on occasion, it renders some bizarre results--dogs with leashes held by invisible masters and ankles without legs appearing in shoes.

In addition, the software only works in urban settings where the pixels blocked by people are on a "dominant planar surface," which makes them simpler to replace. So the program would work fine if a person were walking by a mural of horses grazing in a pasture because the mural in the background would be flat. If the person were walking by real horses grazing in a pasture, the program would be less effective because the background wouldn't be flat.

Photo credit: UCSD Jacobs School of Engineering

Subscribe to the Today in Tech Newsletter

Comments