A Feature-based Method for Shipboard Video Stabilization

Abstract

The aim of this article is to present a digital shipboard video stabilization method to stabilize the unstable video due to the unwanted camera shakes and jitters caused by the motion and vibration of ships. The proposed video stabilization algorithm consists of motion estimation, filtering and compensation. The idea of motion estimation is detecting corner points using adaptive Harris algorithm, then the corner points are tracked by optical flow algorithm to determine the camera motion. In addition, the Fourier series curve fitting is designed to cope with the randomness of unwanted camera motion. After motion filtering, the image sequence is moved to compensate the unwanted camera motion according to the motion filtering results. To evaluate the performance, the video stabilization algorithm is evaluated using the video chips captured on an experimental ship. The results show that the variance of frame motion in X and Y axes has been reduced 85% and 95% respectively. The main limitations are the assumption that there is no scale variance between frames, and the camera motion is described by rigid motion model. Our video stabilization algorithm detects the corner points using adaptive threshold which is based on the image information, the desired corner points are selected and the noise are reduced at the same time. In addition, the motion filter is based on the Fourier series curve fitting method which makes the global motion smoother and restores the ego motion of camera.