Abstract: For large- scale point cloud datasets, traditional point cloud registration methods face the problems of long computation time and significantly reduced registration accuracy. To address these issues, an improved point cloud registration algorithm ISS- 4PCS- ICP (Intrinsic Shape Signatures 4- Points Congruent Sets) is proposed for four- point sets of feature point. Point cloud preprocessing was performed to reduce data size. The ISS (Intrinsic Shape Signatures) method was used to filter out key points with high recognition as input point clouds for the Super 4PCS (4- Points Congruent Sets) method for point cloud coarse registration. In the precise registration iterative nearest point algorithm (ICP) stage, a fusion strategy of KD tree and OC tree was adopted for fast search of nearest neighbor points. The experimental results show that compared with the traditional ICP algorithm and ISS- SAC- IA algorithm, this algorithm has an average calculation time acceleration of about 50%, an average error rate reduction of about 20%, and significantly improves the speed and efficiency of registration.