Abstract: In order to handle fast non-rigid human motion 3D reconstruction based on a single depth sensor,we proposed a non-rigid point cloud registration method base on skeleton coarse alignment. Firstly, we utilizedlocal segmented body part point clouds registration algorithm to correct the skeleton extracted by depthsensor, which effectively improved the accuracy of the skeleton extraction; then, we utilized the skeleton todrive the human body surface to do coarse alignment when there was a large human motion between twoconsecutive frames; lastly, we did fine non-rigid point cloud registration after the coarse alignment to getaccurate reconstruction. The experimental results show that our method can robustly reconstruct fast humanmotion.