Abstract: The point cloud data is uneasy to distinguish and difficult to denoise by outlier 3-D laser scanning. To solve the problems, this paper presents an improved C-means algorithm for solving the 3-D laser scanning point cloud data noise and outliers. The improved C-means algorithm introduces the fuzzy weighting factor that can effectively expand the characteristics of outliers in the dataset and make easier to identify outlier data. The noise is divided into large and small scales in two categories. The C-means clustering algorithm can remove the large scale data smoothing noise and some small noise data using point cloud bilateral filtering method. Compared with the density clustering algorithm, orthogonal total least squares plane fitting and filtering point clouds denoising and feature selection algorithm based on bilateral, the accuracy of denoising is promoted 7.3%, 6.5% and 6%. The experimental results show that the algorithm can remove the noise of large scale, better retention of valid data, improve the effect of denoising.

Key words: C-means, 3D point cloud data, denoising, fuzzy clustering

摘要： 针对三维激光扫描仪采集到的点云数据中离群点不易区分和去噪难度大的问题，提出了一种改进的C均值算法。通过分析三维点云数据特征，在传统C均值算法中引入模糊聚类权重因子，降低类内距离和拉大类间距离，有效增强了离群点特征以降低识别难度。进而将识别出的噪声分类别处理，利用改进的C均值算法去除大尺度噪声，构造双边滤波算法去除小尺度噪声数据。与密度聚类算法、正交整体最小二乘平面拟合和基于特征选择的双边滤波点云去噪等算法相比，去噪准确度分别提升了7.3%、6.5%和6.0%，实验结果表明该算法可以有效去除大尺度噪声并能较好地保留有效数据。

关键词: C均值, 三维点云, 去噪, 模糊聚类