Abstract: The point cloud of curved parts obtained by laser scanning contains noise points， that may affect the surface fitting accuracy of the parts. A denoising and smoothing method based on noise classification was proposed. The kd tree algorithm and K-means clustering algorithm were combined to remove the outlier noise. The curvature of the sampling points was substituted to improve the bilateral filter factor， and the improved bilateral filtering algorithm was used to smooth the non-outlier noise. The proposed method was used to denoise the point cloud of the blade of the guide pulley of the hydraulic torque converter， and compared with the measurement results obtained by coordinate measuring machine. The experimental results showed that the deviations between more than 70% of the measurement points and the denoised point cloud is within ± 10μm. It was found that the proposed method can well preserve the geometric features of the scanned object， and the denoised point cloud can meet the accuracy requirements of the model reconstruction.

Key words: point cloud processing; point cloud denoising; K-means clustering; improved bilateral filtering; hydraulic torque converter