Abstract:3D modelling finds a wide range of applications in robot vision and reverse engineering. However due to the presence of surface scanning noise, accumulative registration errors, and improper data fusion, the reconstructed surfaces from multiple registered range images are often non-smooth and distorted with thick patches, false connections and blurred features. These shortcomings will limit the wide applications of 3D modelling using the latest laser scanning systems. In this paper, the clustering approach, surface segmentation and classification are employed to fuse registered range images to reduce the adverse effect of large accumulative registration errors and heavy scanning noise, minimize the dissimilarity of the fused surface with respect to the original overlapping surfaces, and produce smooth and detailed 3D object computer models. For initialization of the clustering approach, an automatic method is developed, shifting possible corresponding points from different viewpoints toward each other and thus, making sure that the initialization of the cluster centroids are in between the two data sets so that more efficient and effective integration of data can be obtained. Then the principal component analysis (PCA) is employed to segment these centroids into different areas for the representation of the initially fused surface and then the tensor analysis is applied to classify these areas into featured and non-featured ones. In the integration process, the K means and fuzzy c means clustering approaches from the pattern recognition and machine learning literatures are employed to integrate non-featured areas for a smooth fused surface and featured areas for keeping the geometric details. By controlling a parameter, the final integrated surface can be traded off between smoothness and geometric details. Finally the fused point set is triangulated using an improved Delaunay method, guaranteeing a watertight surface. The new method is theoretically guaranteed to converge and minimize the dissimilarity between the final fused surface and original surfaces. A comparative study based on real images shows that the proposed algorithm desirably retains geometric details, produces smooth surface and minimizes the integration error.

Reconstruction of Convex Polytope Compositions from 3D Point-clouds

Generating Large Convex Polytopes Directly on Point Clouds

A clustering approach to free form surface reconstruction from multi-view range images

A Convex Framework for High Resolution 3D Reconstruction.

Reconstruction of Hybrid Models Composed of Analytic and Free-form Surfaces by Using Constrained Optimization Method

Reconstruction of Symmetric Models Composed of Analytic Curves and Surfaces from Point Cloud

Compounded Geometry Image Representation of Point Cloud Surfaces and Its Application

3D Reconstruction Approach for Outdoor Scene Based on Multiple Point Cloud Fusion

A Bicomposition of Conical Projections

A Two-Phase Approach of Progressive Mesh Reconstruction from Unorganized Point Clouds

Hyperbolic-constraint Point Cloud Reconstruction from Single RGB-D Images

Outdoor Scene 3D Reconstruction from Multiple Point Cloud

Plane Fitting in 3D Reconstruction to Preserve Smooth Homogeneous Surfaces

An automatic framework for quadrilateral surface reconstruction with partitions from 3D point clouds

Region segmentation via deep learning and convex optimization

Approximate convex decomposition for 3D meshes with collision-aware concavity and tree search

Scalable Surface Reconstruction from Point Clouds with Extreme Scale and Density Diversity

Concavity-Induced Distance for Unoriented Point Cloud Decomposition

Convex Optimization for Nonrigid Stereo Reconstruction.

Learning Shapes by Convex Composition

Fitting 3D Shapes from Partial and Noisy Point Clouds with Evolutionary Computing