Abstract: Considering the characteristics of unmanned aerial vehicle(UAV) aerial video, a UAV video stabilization method based on sub-block projection is proposed, which improves the accuracy of global motion estimation. This method divides an image into several sub-blocks and wipes off those sub-blocks which contain local motions and are textureless in gray level. Then the motion vectors of the sub-blocks remained are calculated respectively by using gray projection algorithm. Further, the global motion vector is estimated by local motions. Finally, after the motion decision, the tremble motion vector is used for motion compensation. Experiments on real video images of UAV demonstrate that the accuracy of this method is close to the standard motion estimation algorithm, but the time consumption is only one third of it. The result also shows that the effectiveness and the efficiency of the proposed algorithm are superior to the traditional gray projection algorithm.