📜 ⬆️ ⬇️

A graduate student from the United States has created a "pedestrian retoucher" for Google Street View


Recently, only lazy people have not discussed the problem of using personal data from unprotected WiFi networks by Google. The latter, of course, admitted her guilt and promised never again ... never have any particular reason not to believe the company, so many Internet users decided to postpone their own protests until the next time. Of course, besides the above problem, Google has a lot of problems with the use and protection of personal information of various people. For example, such a problem is the Google Street View service. Personally, I, like many other users of the habrasoobshchestva, deeply violet that we can be in the next snapshot taken by Google Mobile. But there are people who are very concerned about this problem.

Of course, after all, the prospect of being in the photo with an unknown wife of his own girlfriend by the hand next to an employee of a competing company is not very happy. There are just paranoids who fear everything and always, including “as if something did not work out”, there are also normal people who really cannot or don’t want to appear on such pictures. The solution of this problem by the corporation was postponed indefinitely, since it is not so easy to remove a person from a photo without spoiling the photo itself (tens of thousands of photos with thousands of people on them).
Why would anyone see the streets of cities with white spots instead of pedestrians? So, the problem has already been solved, or rather, a solution has been found, but has not yet been implemented in the Google Street View system. A graduate student at the University of California has created a “pedestrian retoucher”, a program that automatically replaces people with pieces close to the texture and color scheme of the surrounding landscape. That is, the operation of the system consists of two stages. The first is, in fact, the removal of a pedestrian in the frame. The second is the replacement of the formed void approximating the color and texture of the background surrounding the pedestrian with pieces of graphics. In general, the "retoucher" does not work perfectly, the errors are visible and very well. However, they become apparent if you know where the pedestrian was. But if you just “flip through” the street, then the chance to stumble upon such a retouched place is extremely small. Sometimes you get curious curiosities - for example, you can see a dog with a tight leash that nobody holds, or there you can see sneakers that stand on the sidewalk, “losing” the owner.


The “retoucher” works best with more or less homogeneous landscapes, although it copes well with more complex tasks, for example, removing a person from the background of a street with more details.

In general, the technology is really interesting and in many respects useful. It remains, though a few questions. For example, what to do if a street is photographed, or just an area with a large number of people? Probably, similar situations are difficult for the filter. In any case, you can get acquainted with a more detailed description of the new product by downloading this document here (pdf), where the author of the development describes his brainchild.

And here is the original news.

Source: https://habr.com/ru/post/101253/

All Articles