There are thousands of GoPro-powered time-lapse videos on YouTube, but most of them are a bumpy ride from a camera mounted on top of someone’s head. Microsoft is aiming to transform this type of first-person footage into smoother, less shaky, video with its own hyperlapse work. The software maker has created a method for converting helmet camera activities into hyperlapse videos thanks to its research team. Microsoft’s algorithm reconstructs a scene by using a depth map to create new frames that follow a stable path.
It’s similar to Microsoft’s Photosynth work, with the new frames stitched and blended together to recreate a scene smoothly. The result is a much more stable time-lapse video that effortless glides along frame-by-frame. It’s not always perfect, with some shapes and objects appearing randomly on some of Microsoft’s samples, but most results are impressive. Similar techniques were applied to Google Street View imagery to produce a smooth hyperlapse last year. While Microsoft’s project is simply research for now, the company does promise it’s "working hard" on a Windows app that it hopes to release soon.