The coding method in low bit rate discussed in this paper is a hybrid encoding architecture, which combines global motion compensation and local motion compensation coding. In order to solve the complexity and large quantity of calculations of global motion compensation in this kind of hybrid encoder, the feature based global motion estimation and robust object function and three-level pyramid calculation are proposed to improve the performance of encoder system, especially in speed. This optimal global motion estimation technique enhances the performance of coding system substantially.Some comparison experiments are made in order to validate the effect of the proposed optimization method with different resolutions and esquences in different bit rates.The encoder system speeds up more than 3 times after optimization through comparisons.The proposed method proposed is an attractive technique,which has been accepted by MPEG-4 video encoder optimization group.