三坐标轮廓检测是大口径高次非球面确定性加工过程中的主要面形测量手段。由于原始三坐标数据包含较大的检测误差，无法直接应用于加工过程，本文提出了一组数据处理算法对误差进行全面去除。首先，对获取的检测数据采用基于球心曲面重建的测头半径补偿算法进行测头半径误差补偿，然后对补偿后数据进行坐标系旋转平移误差去除，最后对提取的检测面形残差进行基于KNN的残差噪点过滤。其中，提出的基于球心曲面重建的测头半径补偿算法通过引入一个高精度的测头球心包络面拟合模型，来计算各检测点的测头半径补偿向量，仿真实验证明:算法补偿精度达到RMS<4 nm；提出的基于KNN的残差噪点过滤算法，通过采用插值方法提高样本空间密度和优化噪声度量值的计算，提高了噪点的识别敏感度并实现了噪点的自动化去除。最终根据整个误差清理算法构建了检测点云处理软件，应用实践表明其有效提高了镜面加工过程中检测点云的数据处理精度和效率。

Surface measurement using a Coordinate Measuring Machine(CMM) is the main method of processing large-aperture high order aspheric mirror fabrication. Because three main types of measurement error exist in original data, the achieved surface residual cannot be directly used in mirror fabrication. Therefore, a series of algorithms for cleaning the errors of CMM point clouds is proposed. Firstly, Probe Radius Compensation(PRC) based on aspherical surface reconstruction is used to compensate for probe radius error in acquired data. Then, the compensation data is processed to remove the rotation and translation errors in the coordinate system. Finally, Surface Residual Denoising(SRD) based on KNN is used to denoise the extracted surface residual data. In the PRC algorithm, a high-precision surface fitting model for probe center points is proposed, which takes rotation and translation errors into consideration. With this model, a correcting vector for each point can be calculated to compensate for the probe radius error. The Simulation experiments show that with PRC algorithm, the precision can reach RMS<4 nm. By densifying the sampling points with spatial interpolation and optimizing the noise characterization, the proposed SRD algorithm can identify the noise points with a high degree of sensitivity, making the denoising intelligent. Software was constructed according to the error cleaning algorithm and its application shows that it can effectively improve the accuracy and efficiency of CMM point cloud processing during aspheric mirror fabrication.