This paper addresses the limitations of traditional point cloud registration methods, including low computational efficiency, sparse feature point extraction, and weak robustness, by introducing a novel fast point feature histogram (FPFH) closest distance ratio point cloud registration method. First, the original point cloud data is downsampled by voxels, and the feature descriptors are obtained from the downsampled data using the FPFH method. Then, the normal vector threshold of the bidirectional K dimension tree and the nearest neighbor distance ratio method are used to extract features, change the distance weight, count more point pair parameters, eliminate mismatched point pairs, speed up the point pair search, and make the registration more accurate. Huber loss function weight is introduced to optimize the objective function of rigid body registration model to enhance robustness; Then the sample consensus based method (SAC-IA) is used for initial coarse registration, and the iterative closest point (ICP) point to surface method is used to complete fine registration; In order to avoid the local minimum, Levenberg-Marquardt (LM) algorithm and singular value decomposition (SVD) alternate iteration strategy are used for optimization. Compared with ICP, SAC-IA+ICP, Intrinsic Shape Signatures (ISS)+ICP and FPFH+ICP, the experiment shows that the registration speed of the proposed method is 31.07%, 53.46%, 3.09% and 25.05% higher than that of the ISS, 49.09% higher than that of the ISS, and an order of magnitude higher than that of ICP and SAC-IA+ICP. When the number of iterations is small and the precision setting is large, the registration precision is the same as that of the FPFH+ICP method, but the speed is faster than that of the FPFH method. In the case of noise, the speed of this algorithm is 30.53% higher and the registration accuracy is 15.47% higher than that of FPFH, which is smoother; Compared with ISS registration algorithm, the registration accuracy is improved by an order of magnitude, and the average registration time is slightly faster; ICP and SAC-IA+ICP methods failed to register and did not participate in the comparison. The experimental results show that this method is feasible, the registration speed and accuracy are improved, and the robustness and stability are better.