Google algorithm creates video from individual Street View photos

Scientists at Google have developed an algorithm that can create smooth video images from individual photos in Google Street View. In a demo video, the scientists show what the 24fps images will eventually look like and which Street View photos they are based on.

Google’s algorithm has a simple principle. The scientists take a series of Street View images, delete a frame, and then have their algorithm reconstruct that frame. Then they have the algorithm compare the reconstructed frame with the original frame, from which the algorithm can learn lessons. For example, a series of images of approximately 2 fps is made into a series of 25 fps.

Google’s scientists had to use 100,000 sets of photos from Google Street View to get the algorithm, called DeepStereo, sharp enough to deal with things like obscured fields of view due to obstacles. “Ultimately, you get results that are hard to distinguish from the original images,” Google researcher John Flynn told MIT Technology Review.

For now, users will not be using a video version of Street View yet. At the moment it takes about 12 minutes to have one frame put together by the algorithm. The Google team is therefore still working on further optimizing the algorithm. Flynn has also written and published an account of the work to date.