Aiming at problem of large joint error of two groups of point cloud models with large angular transformation caused by incomplet inclusive relationship

an improved iterative closest point（ ICP） registration algorithm is proposed. Normal size of in the point cloud data is calculated

and the point with bigger change is extracted as a pre-reference point. The point cloud registration strategy is optimized to get the rough registration model. Based on this model

the corresponding pairs of curvature and dot pitch are constrained

and the exact point pairs are obtained

point cloud data can be joint accurately. The experimental results show that the proposed algorithm can be used to joint the model with large difference in angle and can not only be able to complete the integration of the common part of the two groups

but also can retain data that not comparable between the two groups of point clouds completely.