| 17 | 0 | 28 |
| 下载次数 | 被引频次 | 阅读次数 |
三维点云的原始点在扫描过程中通常会受到传感器测量误差或环境等因素的影响而产生噪声。点云中的噪声会影响目标检测等下游任务，因此开发有效的去噪算法具有重要意义。目前主流的去噪算法一般采用人工合成的数据集进行训练，因此一般在特定数据集上表现良好，但在复杂度较高的真实数据集上的泛化能力有限。针对以上问题，提出了一种基于噪声模型和概率分布理论的自监督点云去噪算法(PSSNet)。该算法在已知噪声分布的单噪声模型的假设下，通过添加同类型噪声干扰生成新的噪声样本，然后使用新生成噪声样本与原始噪声样本2个点云集合进行配对训练，再将网络的输出进行二次处理即可得到去噪点云。PSSNet分别在合成数据集和真实数据集上单独进行训练并测试。结果显示，PSSNet在评价指标上可以达到与有监督去噪方法相近的水平。
Abstract：The raw points of 3D point clouds are often contaminated by noise during the scanning process due to factors such as sensor measurement errors or environmental interference.Noise in point clouds can adversely affect downstream tasks such as object detection, making the development of effective denoising algorithms highly significant.Currently, mainstream denoising algorithms are typically trained on synthetic datasets, which often leads to strong performance on specific datasets but limited generalization ability on more complex real-world datasets.To address these issues, this paper proposes a self-supervised point cloud denoising algorithm named PSSNet, based on noise modeling and probabilistic distribution theory.Under the assumption of a single noise type with known distribution, the algorithm generates augmented noisy samples by adding the same type of noise perturbation.These newly generated noisy samples are then paired with the original noisy samples for training.After secondary processing of the network output, the denoised point clouds can be obtained.PSSNet is separately trained and tested on both synthetic and real-world datasets.The results demonstrate that PSSNet achieves performance comparable to supervised denoising methods in terms of standard evaluation metrics.
[1] 王嘉鑫，赵夫群.点云数据预处理研究[J].现代信息科技，2020,4(2):129-130.
[2] LEHTINEN J,MUNKBERG J,HASSELGREN J,et al.Noise2noise:learning image restoration without clean data[C]//International Conference on Machine Learning.Stockholm:PMLR,2018:2965-2974.
[3] WU Z,SONG S,KHOSLA A,et al.3D shapenets:a deep representation for volumetric shapes[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Boston:IEEE Press,2015:1912-1920.
[4] YU L,LI X,FU C W,et al.Pu-net:point cloud upsampling network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE Press,2018:2790-2799.
[5] SERNA A,MARCOTEGUI B,GOULETTE F,et al.Paris-rue-madame database:a 3d mobile laser scanner dataset for benchmarking urban detection,segmentation and classification methods[C]//4th International Conference on Pattern Recognition,Applications and Methods.Angers:SUTEPRESS,2014:129-136.
[6] HERMOSILLA P,RITSCHEL T,ROPINSKI T.Total denoising:unsupervised learning of 3D point cloud cleaning[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.Seoul:IEEE Press,2019:52-60.
[7] MATTEI E,CASTRODAD A.Point cloud denoising via moving RPCA[J].Computer Graphics Forum,2017,36(8):123-137.
[8] ZENG J,CHEUNG G,NG M,et al.3D point cloud denoising using graph Laplacian regularization of a low dimensional manifold model[J].IEEE Transactions on Image Processing,2019,29:3474-3489.
[9] ACHLIOPTAS P,DIAMANTI O,MITLIAGKAS I,et al.Learning representations and generative models for 3D point clouds [C]//International Conference on Machine Learning.Stockholm:PMIR,2018:40-49.
[10] JOHNSON J,RAVI N,REIZENSTEIN J,et al.Accelerating 3d deep learning with pytorch3d [J].SIGGRAPH Asia 2020 Courses.New York:ACM,2019:1-3.
基本信息:
DOI：10.16426/j.cnki.jcdzdk.2026.02.015
中图分类号:TP391.41;TP18
引用信息:
[1]王坤鹏.基于噪声模型和概率分布理论的自监督点云去噪算法[J].舰船电子对抗,2026,49(02):85-91.DOI:10.16426/j.cnki.jcdzdk.2026.02.015.
2026-04-25
2026-04-25