Abstract: In recent years, more and more people use hand-held devices to shoot videos to express themselves or social problems, etc. This kind of video often has the phenomenon of jitter, how to eliminate the random jitter in the video sequence has become a research focus in recent years.In this paper a video stabilization algorithm based on Savitzky-Golay filtering and L1 norm optimization was proposed for real time application,which is used to dejitter video captured by hand-held cameras. Firstly,the motion is estimated based on the feature points of inter-frame pairing. Next,Savitzky-Golay filtering algorithm is used for path smoothing to remove the high frequency jitter components in the moving path, so as to avoid the influence of noise on the subsequent generation of stable paths.Then, the L1 norm optimization algorithm is used to optimize the smooth effective path information to fit the final stable path. Finally, the motion compensation of the original video image sequence is carried out to generate a stable image sequence. The compared experimental results demonstrate that the proposed method is better than the original L1 norm optimization algorithm in image stabilization and faster than the current two-dimensional image stabilization algorithm.