Google lands patent for automatic object recognition in videos

 

Google has already been working on patents that could pick out faces and song melodies in our YouTube clips. Now, it might just have the ultimate tool: the technique in a just-granted patent could pick out objects in a video, whether they're living or not. Instead of asking the creator to label objects every time, Google proposes using a database of "feature vectors" such as color, movement, shape and texture to automatically identify subjects in the frame through their common traits -- a cat's ears and fast movement would separate it from the ball of yarn it's attacking, for example. Movie makers themselves could provide a lot of the underlying material just by naming and tagging enough of their clips, with the more accurate labels helping to separate the wheat from the chaff if an automated visual ranking system falls short. The one mystery is what Google plans to do with its newfound observational skills, if anything, although the most logical step would be to fill in YouTube keywords without any user intervention -- a potential time-saver when we're uploading that twelfth consecutive pet video.