A robust video stabilization algorithm based on feature matching is put forward. First,the traditional KLT method is improved by introducing an adaptive illumination independent model to realize the robust feature tracking. Then the feature matching errors and motion consistence principle are used to improve the feature reliability. Finally,the correspondence between the matched features is used to build an affine motion model to realize image stabilization. Experiment results show that the proposed method is robust to both moving object and illumination change.