4790 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 54, NO. 8, AUGUST 2016 Robust Segmentation for Large Volumes of Laser Scanning Three-Dimensional Point Cloud Data Abdul Nurunnabi, David Belton, and Geoff West, Senior Member, IEEE Abstract—This paper investigates the problems of outliers and/or noise in surface segmentation and proposes a statistically robust segmentation algorithm for laser scanning 3-D point cloud data. Principal component analysis (PCA)-based local saliency features, e.g., normal and curvature, have been frequently used in many ways for point cloud segmentation. However, PCA is sensitive to outliers; saliency features from PCA are nonrobust and inaccurate in the presence of outliers; consequently, segmentation results can be erroneous and unreliable. As a remedy, robust techniques, e.g., RANdom SAmple Consensus (RANSAC), and/or robust versions of PCA (RPCA) have been proposed. However, RANSAC is influenced by the well-known swamping effect, and RPCA methods are computationally intensive for point cloud pro- cessing. We propose a region growing based robust segmentation algorithm that uses a recently introduced maximum consistency with minimum distance based robust diagnostic PCA (RDPCA) approach to get robust saliency features. Experiments using syn- thetic and laser scanning data sets show that the RDPCA-based method has an intrinsic ability to deal with outlier- and/or noise– contaminated data. Results for a synthetic data set show that RDPCA is 105 times faster than RPCA and gives more accurate and robust results when compared with other segmentation meth- ods. Compared with RANSAC and RPCA based methods, RDPCA takes almost the same time as RANSAC, but RANSAC results are markedly worse than RPCA and RDPCA results. Coupled with a segment merging algorithm, the proposed method is efficient for huge volumes of point cloud data consisting of complex objects sur- faces from mobile, terrestrial, and aerial laser scanning systems. Index Terms—Feature extraction, object modeling, outlier, re- gion growing, robust normal, robustness, segmentation, surface reconstruction. I. I NTRODUCTION S EGMENTATION is an actively researched task for many applications, including object shape recognition, model- ing and geometry analysis, surface reconstruction and feature extraction in computer vision, pattern recognition, photogram- metry, remote sensing, and robotics [1]–[4]. It is a process of classifying and labeling data points into a number of separate Manuscript received April 11, 2015; revised November 26, 2015; accepted February 27, 2016. Date of publication May 5, 2016; date of current version June 1, 2016. This work was supported in part by a Curtin University Inter- national Postgraduate Research Scholarship through a Ph.D. research and in part by a top-up scholarship from the Cooperative Research Centre for Spatial Information, whose activities are funded by the Australian Commonwealth’s Cooperative Research Centres Programme. A. Nurunnabi and G. West are with the Department of Spatial Sciences and the Cooperative Research Centre for Spatial Information, Curtin University, Perth, W.A. 6845, Australia (e-mail: abdul.nurunnabi@postgrad.curtin.edu.au; g.west@curtin.edu.au). D. Belton is with the Department of Spatial Sciences, Curtin University, Perth, W.A. 6845, Australia (e-mail: d.belton@curtin.edu.au). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2016.2551546 groups/regions, each corresponding to the specific shape of a surface of an object. It is important in processing laser scanning (LS) point cloud data (PCD), and the success of information re- trieval largely depends on the quality of the results [5]. Segmen- tation in PCD is not trivial, because the three-dimensional (3-D) georeferenced x, y, z points are usually incomplete, sparsely populated, and unorganized (i.e., no information about connect- ing neighboring points), with no knowledge about the statistical distribution of the points, as well as point density variation. Moreover, the presence of different types of outliers and noise makes the segmentation process complicated and challenging. Complex topology and the presence of sharp features (e.g., edges and corners) further exacerbate the complexity. Region growing based segmentation is a most commonly used segmentation approach. Principal component analysis (PCA) has been widely used to estimate local saliency features (SFs) such as normal and curvature that are used in region growing [6]–[8]. Mitra et al. [9] showed that the sensitivity of PCA to outliers means that it fails to accurately fit planes and the resultant plane parameters are unreliable, nonrobust, and misleading. Outliers and/or noise can cause several problems, including the tendency to smooth sharp features. Hence, seg- mentation results can be inaccurate and nonrobust. Li et al. [10] pointed out that if correct normals are robustly estimated for each point, the geometry of corrupted point clouds can be well determined. Nurunnabi et al. [11], [12] showed that robust and diagnostic statistical approaches reduce the influence of outliers/noise on PCA, produce robust SFs, and can be used for robust segmentation. This paper introduces two algorithms: a region growing based robust segmentation algorithm for 3-D PCD that can be generated by mobile laser scanning (MLS), terrestrial laser scanning (TLS), and aerial laser scanning (ALS) systems, which is an extension of the method proposed recently in [13], and a merging algorithm to combine the segments of sliced data that cannot be processed as one data set. The methods use robust diagnostic PCA (RDPCA) as an alternative of the robust PCA used in [13] to reduce outlier influence on the estimated lo- cal SFs for accurate and robust segmentation. The segmentation algorithm is efficient for both planar (e.g., building facades and roofs) and nonplanar (e.g., cylindrical objects such as sign and light poles) complex objects, but objects such as trees cannot be segmented as one object since estimating surface orientation and region growing is not trivial for leaves and branches. Using RDPCA significantly reduces the segmentation time than using the robust PCA based method [13]. We show the accuracy, efficiency, and robustness of the new algorithms for segmenting artificial and real PCD in the presence of outliers/noise and sharp features. 0196-2892 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.