Video Stabilization with SIFT
Katie Zutter Olivia Zhao Sam Erickson
Whether with a cell phone or another device, videos taken without a professional camera are often shaky and unstable. Even with high-performance sports cameras, videos taken under certain circumstances can lead to unsatiated products. Professional cameras may have hardware that either prevents the camera from shaking or measures how the camera is shaking with high-end sensors, but these features make consumer cameras too costly or bulky.
Many professional cameras come standard with software that executes stabilization algorithms. This project will explore one such stabilization algorithm: the use of Scale Invariant Feature Transform (SIFT) to track the motion of features between consecutive video frames. The goal is to implement the SIFT algorithm in such a way that it automatically adjusts frames to create one smooth, seamless video. Because not all movement between frames is due to camera shakes, our implementation must be able to discern the difference between intentional and accidental motion.
We capture videos of important moments of our lives; image stabilization ensures that after the moment passes we are not left with a blur of footage. Solving this problem would mean that all camera users could take videos without worry of camera shake or random errors, no matter what kind of camera is within their budget. Advancing image stabilization also means that older adults and people with disabilities that affect fine motor control can more confidently capture video.