This paper proposes a point cloud registration algorithm based on extended point feature histogram (EPFH) feature. In the proposed algorithm, the strategy of rough registration prior to fine registration is adopted. The purpose of this paper is to overcome the problems of low registration accuracy and slow speed which are encountered in traditional registration methods. Initially, the intrinsic shape signature (ISS) feature-detection algorithm is used to obtain the salient feature-point set on the point clouds. Then, the EPFH feature description is applied on these feature points. Subsequently, the rigid body transformation matrix is estimated using the sampling consistency algorithm to complete the initial registration of the point clouds and the target point clouds. The k-d tree-based iterative nearest-point algorithm is used to implement the fine registration of the two-point clouds. Finally, experimental verification is performed by applying the proposed algorithm to the public data set and the terracotta warrior data set. The experimental results show that the proposed algorithm exhibits higher registration accuracy and higher speed than traditional methods.