An adaptive temporal compressive sensing for video based on signal correlation is proposed, which can judge the motion of the object adaptively and reconstruct the signal targeted in the process of the super time resolution video imaging. The proposed method separates the observed image into regions of different amount-of-movement, and then reconstructs these regions with targeted dictionaries, which are trained from corresponding video samples. In the process of video reconstruction, block reconstruction of the coded exposure video is being done. This fast reconstructed video is used to compute the correlation coefficients between the neighbor frame image blocks. Local motions are then estimated by the correlation coefficients, and finally, the dictionaries can be adaptively selected according to the motion information to reconstruct the video. Simulation results show that the proposed method can obtain the motion distribution accurately, and the quality of the reconstructed video is increased while the reconstruction time is reduced.