3D point cloud denoising method based on improved PCPNet

Aiming at the obtained 3D point cloud data, there are a large amount of noise and the loss of local details in the process of data processing, an improved deep learning network based on PCPNET was proposed for effectively denoising. In the encoder, the feature enhancement module is used to fully explore the potential local geometric features of the noisy point cloud, process the point clouds of different scales, and design the corresponding loss function to feed back the network, optimize the network performance, and achieve the purpose of point cloud denoising, which is conducive to improving the registration accuracy of 3D point cloud in the future. Experimental results show that compared with traditional methods, the proposed method has better denoising effect, has a certain robustness to different noises, and can effectively retain local edge information.