科学美国人60秒:Software Finds Best Parts of Boring Video

Your New Year’s resolution to start recording every single thing you do, a.k.a. lifelogging, seemed like a fun idea. But now even you don’t want to sit through all the footage of last week’s family barbeque to figure out how the dog got out of the yard.
 
Thankfully, you won’t have to. You’ll run a computer program that evaluates hours of insipid images and automatically culls together the most interesting moments into a personalized trailer.
 
At least that’s what machine-learning researchers at Carnegie Mellon University are promising with LiveLight. It’s an automated system for summarizing unedited video. LiveLight’s algorithm can analyze footage from Google Glass, for example, and pick out unique activity while skipping over parts that are repetitive or where nothing happens.
 
The researchers see several uses, such as analyzing long stretches of video from surveillance and traffic cameras to find an intruder or the cause of an accident. It could also be an alternative to video search engines, something computer scientists have been working on for years without much success.
 
Google helped fund the development of LiveLight. Let’s hope they make it a one-click option on all YouTube videos.
 
—Larry Greenemeier
 
(The above text is a transcript of this podcast)