Abstract: Decreasing the random motion while shooting video timely is the main purpose by electronic imagestabilization （EIS） algorithm. We proposed a new algorithm that would improve speed and accuracy.Firstly, self-adaptive extracting feature points are realized; then, three-point and one-point correction methodare used that meet the convex optimization conditions to solve the affine matrix; after that, each step of thealgorithm process is controlled and optimized so that the errors are minimized in the whole procedure; at last,local motion object’s area is extracted to realize a dynamic compensation, thus the time is reduced from 20 msto 1 ms in this step. Experimental results demonstrate that the proposed algorithm processing speed can reachat 7 ms per frame （excluding the feature-pints extract process）, and work well for all the videos used in theexperiment.