An adaptive electronic image stabilization algorithm was proposed to remove jitter from videos with foreground moving object.Visual attention on motion was used to make fast foreground segmentation.It made difference between three adjacent frames based on blocks,and the temporal-spatial consistency was validated to label foreground moving object.The grid-sampling and significant degree sorting were proposed to detect global distinct features in the background area of reference and match in current frame.The distance criterion based on statistical distribution was used to delete mismatched points,which improved the speed and accuracy of global motion estimation without iteration process.Then Sage-Husa adaptive filtering method was described and improved,which corrected the process and observation noise by estimating their statistical property in real time.The low-frequency visual smooth as uniform motion was simulated.Thus,it can remove camera jitter and retain camera scanning simultaneously.With VC++ implementation and a computer(Intel Core(TM)2,2.33 GHz),processing image of size 320×240 reached 22 fps speed.The experimental results illustrate that the proposed algorithm can stabilize the inter-frame jitter with large or multi moving targets and track the real scene with smooth visuality.