Software video stabilization in a fixed point arithmetic

Abstract

This paper discusses a new approach to the stabilization of video sequences. The input of a video stabilization system is a video sequence of relatively static scene, which is ragged due to a mechanical shaking of the camera. This type shaking is often amplified by a camera zoom. Because of this, the video stabilization algorithms are often useful in combination with the security monitoring systems, which often must deal with the zoomed scenes. The video stabilization means a correction of such video sequences, so the stabilized output appears as steady. Our algorithm is based on the fixed-point arithmetic. For this reason it is suitable for use on devices that are not optimized for fixed-point arithmetic, like the DSP architectures and other embedded systems.