保持特征的散乱点云数据去噪

发布时间：2018-05-29 16:31

本文选题：点云去噪 + 加权模糊C均值　；参考：《光学精密工程》2017年12期

【摘要】：为保证在去除点云数据噪声的同时不损失模型的细节特征,提出了一种基于特征信息的加权模糊C均值聚类去噪算法。首先,构建点云K-D树拓扑结构,根据点的r半径球邻域点统计特性去除大尺度离群噪声点。然后,利用主元分析法估算点云的曲率和法向量,根据曲率特征标识点云数据的特征区域,并采用特征加权模糊C均值聚类算法对特征区域去噪,采用加权模糊C均值聚类算法对非特征区域去噪。最后,使用双边滤波器对点云模型进行平滑。对提出的算法进行了验证实验,结果显示:去噪后点云模型的最大偏差保持在模型尺寸的0.15%以内;标准偏差保持在模型尺寸的0.03%以内。本文算法能够在有效去除不同尺度和强度的噪声的同时不损失点云模型的细节特征,去噪精度高,且对不同的噪声模型具有较强的鲁棒性。
[Abstract]:A weighted fuzzy C-means clustering denoising algorithm based on feature information is proposed to ensure that the point cloud data noise is removed without losing the detailed features of the model. Firstly, the point cloud K-D tree topology is constructed, and the large scale outliers are removed according to the statistical characteristics of the r-radius spherical neighborhood points. Then, the curvature and normal vector of point cloud are estimated by principal component analysis, the feature region of point cloud data is identified according to curvature feature, and the feature region is de-noised by feature weighted fuzzy C-means clustering algorithm. A weighted fuzzy C-means clustering algorithm is used to Denoise non-feature regions. Finally, a two-sided filter is used to smooth the point cloud model. The experimental results show that the maximum deviation of the de-noised point cloud model is within 0.15% of the model size, and the standard deviation is less than 0.03% of the model size. The proposed algorithm can effectively remove noise of different scales and intensities without losing the detailed features of the point cloud model. It has high denoising accuracy and is robust to different noise models.
【作者单位】：北京宇航系统工程研究所;
【基金】：国家国际科技合作专项资金资助项目(No.2013DFA51360)
【分类号】：TP391.41

【相似文献】