A large-scale rapid DEM mosaicking method was proposed. In this method， dense tie-points were extracted from the DEM overlap areas using a block matching strategy， and the systematic errors of plane and elevation were corrected through block adjustment. Thus， an entire DEM set was generated using the mosaicking strategy of first along-track followed by cross-track. A computing strategy was adopted to accelerate the processing speed to improve the efficiency of the entire mosaicking process. The experimental results showed that the CPU/GPU asynchronous parallel method proposed in this study improves the processing efficiency by more than 18 times that of a single CPU， and realises the seamless mosaicking of multi-stripe DEMs.