To solve the problem that the multi scale mixed noise of point cloud data obtained by 3D laser scanner seriously affects the subsequent 3D model reconstruction,a point cloud denoising algorithm based on improved density peak clustering algorithm(DPC)and feature partition is proposed.Firstly,the improved DPC algorithm is used to remove the large scale noise far away from the main body of the point cloud.Then,principals of component analysis(PCA)and surface variation are used to obtain the normal vector and curvature information of the point cloud.At the same time,neighborhood propagation method is adopted to adjust the normal vector direction and divide the point cloud according to the curvature.Adaptive bilateral filtering and orthogonal total least squares plane fitting are applied to smooth and denoise the point cloud in feature area and flat area respectively.The experimental results show that under the bunny and block model with mixed noise,the maximum error of the point cloud data is 0.235mm and 0.157mm respectively,and the average error is 0.029mm and 0.009mm respectively,which can achieve good denoising effect and reduce the complexity of parameter setting.