Video Stabilization Using SIFT Features, Fuzzy Clustering, and Kalman Filtering

Video stabilization removes unwanted motion from video sequences, often caused by vibrations or other instabilities. This improves video viewability and can aid in detection and tracking in computer vision algorithms. We have developed a digital video stabilization process using scale-invariant feature transform (SIFT) features for tracking motion between frames. These features provide information about location and orientation in each frame. The orientation information is generally not available with other features, so we employ this knowledge directly in motion estimation. We use a fuzzy clustering scheme to separate the SIFT features representing camera motion from those representing the motion of moving objects in the scene. Each frame's translation and rotation is accumulated over time, and a Kalman filter is applied to estimate the desired motion. We provide experimental results from several video sequences using peak signal-to-noise ratio (PSNR) and qualitative analysis to demonstrate the results of each design decision we made in the development of this video stabilization method.