Compressed Video Sensing (CVS) has great significance to the scenarios with a resource-deprived video acquisition side. Reconstruction algorithm is the key technique in compressed video sensing. The Multi-Hypothesis (MH) prediction based “prediction-residual reconstruction” framework has good reconstruction performance. However, most of the existing multi-hypothesis prediction algorithms are proposed in measurement domain, which cause block artifacts in the predicted frames and decrease reconstruction accuracy due to the restriction of non-overlapping block partitioning. To address this issue, this paper proposes a two-stage Multi-Hypothesis Reconstruction (2sMHR) idea by incorporating the measurement-domain MH prediction with pixel-domain MH prediction. Two implementation schemes, GOP-wise (Gw) and Frame-wise (Fw) scheme, are designed for the 2sMHR. Simulation results show that the proposed 2sMHR algorithm can effectively reduce block artifacts and obtain higher video reconstruction accuracy while requiring lower computational complexity than the state-of- the-art CVS prediction methods.