Google has developed a technique that mimics professional camera movement and applied it to videos recorded by handheld devices, such as smartphones. While professionals use tripods and dollies, panning a swiftly running cheetah at the San Diego Zoo and getting a clear picture could become quite frustrating on an Android or an iOS phone.
The technology supports an algorithm that automatically determines the best camera path and recasts the video as if it were filmed using stabilization equipment. It's part of the research at Google looking at methods to make casual videos appear more professional. The research is being integrated into YouTube to support videographers.
In a research paper -- Auto-Directed Video Stabilization with Robust L1 Optimal Camera Paths – Google researchers describe the process of dividing the original, shaky camera path into segments to create a smooth video. The optimization finds the best of all possible partitions, using what researchers call a computationally stable algorithm.
More recently, the research has been working to resolve a related problem common in videos shot from mobile phones. Camera sensors in phones contain an electronic rolling shutter. When taking a picture the image is not captured instantaneously, but rather one row of pixels at a time, with a small delay when going from one to the next.
Consequently, if the camera moves during the capture it causes image distortions that are especially noticeable in videos where the camera shake is independent across frames. Google researchers are working to smooth out the shakiness of the video.
Microsoft Research also continues to work with video, but for the patent company headquartered in Redmond, Wash., the work points to mobile gaming. Not the type of gaming where the keyboard becomes the weapon that slingshots at "Angry Birds," but mobile motion that turn the phone into a sword. With phone in hand, the users block the attacks of the counterpart in the game.
Phone-to-phone mobile motion games must have the ability to calculate accurate distance and range from each other. Then range, speed, and accuracy are calculated. The process works similar to Kinect, a fixed infrastructure motion capture system that supports game motion in real-time.