Abstract In the process of collecting data from point clouds, some factors, such as precision equipment, operator experience, environmental conditions, as well as characteristic diffraction effects of electromagnetic waves, changes in surface characteristics of test objects, and the impact of integration processes, are involved data and records that may display some point noise in the data point cloud. In addition to noise caused by measurement errors, practical applications can be affected by external interference and obstacles, which causes deviating values in the data point cloud to differ from the measurement entity. A 3D scanner, represented by the depth of the camera, is used to collect data. To eliminate these noises and anomalies, it is necessary to filter the initial points of the cloud. In this article, we suggest using statistical filtering algorithms to perform statistical analysis on each point community, remove points that do not meet predetermined standards, and remove outliers. To verify the effectiveness of the algorithm, it is compared with other filtering algorithms. The experimental results show that the filter algorithm proposed in this thesis has better effects on outlier points and can obtain the general shape properties of the point cloud.