<title>Image sequence stabilization by low-pass filtering of interframe motion</title>

Abstract

ABSTRACT An image sequence stabilization system that removes translational jitter while preserving intentional camera pan ispresented. The video sequence is processed to acquire global camera translations from frame to frame (global interframemotion vectors) by motion estimation. The resulting motion vectors are accumulated to construct an absolute frame positionvs. frame number signal. This signal is low-pass filtered to remove high frequency components caused by jitter, and retainlow frequency parts representing the intentional camera pan. Correction vectors for image frames are obtained bysubtracting the absolute frame position from the low-pass filtered value, and stabilization is achieved by the correspondingtranslation of image frames.Keywords: Image sequence stabilization, Jitter compensation. 1. INTRODUCTION Image sequence stabilization (ISS) is the process of removing unwanted camera motion from an image sequence so as togenerate a compensated video [1-17]. While some applications such as motion analysis and structure-from-motionalgorithms require the entire global motion to be removed, this task is rather referred to as motion compensation. Imagestabilization is primarily concerned with the elimination of undesired image fluctuations that are usually caused byinstabilities of the imaging system. Image sequences can display jiggles caused by vibrations of the operator hand for videoacquired by hand-held camcorders, or oscillations in the platform position if the camera is mounted onto a moving platform.The image stabilization task can be subdivided into two basic systems: 1) the motion estimation system 2) the motioncorrection system. Accordingly ISS techniques can be classified into three categories:i. Mechanical-Optical systems: Vibration feedback is achieved via gyro sensors [2] and the refraction angle is changedaccording to camera inclination [3].ii. Mechanical-Digital systems: The absolute vertical direction is tracked by a sensor [4] and movement correction isperformed by digital processing.iii. Fully Digital systems: Global motion estimation and movement correction are both performed by digital processingunits [5-17].Recent research is concentrated in digital image stabilization (DIS) systems because of the low-cost and lightweightimplementation ability.Most of the research in the field of image stabilization is mainly concentrated in the motion estimation part of the system.Global motion estimation by processing local motion vectors of subimages obtained by block-matching has been proposedin [5]. Motion estimation based on edge pattern matching is demonstrated in [6]. A multiresolution iterative motionestimation scheme that estimates affine motion parameters between levels of the Laplacian pyramid images is described in[7]. A pyramidal hardware system using mosaic based registration is presented in [8]. Morimoto and Chellappa describe atwo-dimensional feature-based multiresolution motion estimation system [9]. Motion estimation using bit-plane matchingand gray-coded bit-plane matching is demonstrated in [10] and [11] respectively. An ISS system making use of phasecorrelation based motion estimation has been demonstrated in [12]. Yao et al. presents a technique to compensate for 3-D