A robust UAS video stabilization method with hierarchical clustering

Video stabilization is an essential component on unmanned aircraft systems (UAS) with the ability to provide a steady perspective for aerial surveillance. Traditional UAS video stabilization methods compensate each frame by utilizing affine model and taking every jitter as a global transformation, which is generally effective when UAS loads CCD sensor camera and UAS flies high enough to assume that the whole picture is in a same plane. However, if these two conditions are not fulfilled, traditional methods may suffer constant wobble and artifact. In order to stabilize UAS surveillance videos in more general cases, spatial and temporal context should be incorporated. In this paper, we present a robust method for UAS video stabilization adaptive to various scenarios. Since each image can be divided into many small planes, we introduce hierarchical clustering to distinguish them and smoothing them separately for better robustness. Then our motion compensation method utilizes multiple grids transformation to eliminate uneven jitter, which is effective on low altitude and rolling shutter. Experiments provide a comparison between our approach and traditional approaches, demonstrating the advantages of our method in challenging scenarios.