Image fusion is a process to combine multiple frames of the same scene into one image. The popular image fusion methods mainly concentrate on static image fusion and lack spatial-temporal adaptability. The conventional multi-resolution image fusion algorithms have not fully exploited the temporal information. To resolve this problem, we present a novel dynamic image fusion algorithm based on Kalman filtered compressed sensing. The fusion procedure characterized by estimation fusion is completed in state space. A parametric fusion model is proposed to learn and combine spatial and temporal information simultaneously. The experiments on the ground-truth data sets show that the proposed fusion algorithm offers a considerable improvement on the dynamic fusion performance and rivals the traditional multi-resolution-based fusion methods.