In the process of reverse engineering, due to the influence of instrument accuracy, human disturbance, complex environment, and other factors, there will be a large number of noise points and uneven distribution using laser scanning equipment to obtain point cloud data. This will have a certain impact on subsequent point cloud processing and model reconstruction. A point cloud denoising algorithm combining statistical filtering and spatial density clustering algorithm (DBSCAN) is proposed for this purpose. This algorithm firstly establishes the topological relationship of the point cloud through the K-dimension (kd tree) to accelerate neighborhood searching speed and complete coarse denoising of the point cloud data. Secondly, by improving the neighborhood radius selection method of the density clustering algorithm, precise denoising of the point cloud data is achieved. At last, the point cloud in the scanning point cloud database of Stanford University and the measured machine element are selected for experimental verification. The experimental results show that, compared with a single bilateral filtering and radius filtering algorithm, this method can remove irrelevant noise while retaining the original geometric features of objects and meet the subsequent modeling accuracy.