LiveLight uses machine learning and algorithms to automatically edit video reels into a montage of the most interesting clips.
It was once the case that only professional film editors knew what it was like to sift through hours of footage to find the best content. With smartphones and consumer-level cameras now enabling anyone to shoot as much video as they like, the job of editing it down into a presentable format is familiar to many. In the past we’ve seen apps such as Cinch make the process possible on smartphones themselves. Now researchers at Carnegie Mellon University in Pennsylvania have created LiveLight, which uses machine learning and algorithms to automatically edit video reels into a montage of the most interesting clips.
Computer scientists at the college developed a method of scanning footage for elements typically deemed uninteresting, such as limited action and repetitive shots. According to the team, the algorithm compiles a “learned dictionary” of events from each piece of footage as it scans, which compares sequences and decides which ones to discard and which ones to keep. For example, a long static shot of passing traffic would be reduced to a short clip because of its repetitive nature, but a crash in the same footage would be identified as a new event in its dictionary and flagged for inclusion in the summary.
Watch the video below for a demonstration of the program:
LiveLight is useful for those who don’t want to spend hours editing long pieces of footage, but could also provide security teams with summaries of unusual activity over periods of time without having to sit through the entire raw video. At the same time, those shooting footage who want to upload it to the web using mobile data can cut down their file sizes using the software, meaning quicker upload times and reduced data charges.
The team envisions the arrival of products such as Google Glass to result in the creation of yet more unedited user produced content, and hopes LiveLight will provide the tools to quickly break it down into useful clips. Are there other ways to help amateur filmmakers create better video content?