A Clustering Approach to Free Form Surface Reconstruction from Multi-view Range Images

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 error, 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, we first employ the principal component analysis and hierarchical segmentation to classify surface patches into featured and non-features areas, the k-means clustering and fuzzy c means clustering approaches from the pattern recognition literature are then employed to integrate these areas respectively. Finally the fused point set is triangulated using the improved Delaunay method, guaranteeing a watertight surface. The new method is theoretically guaranteed to converge and combine the advantage of exclusive k means clustering and overlapping fuzzy c means clustering methods. A comparative study based on real images shows that the proposed algorithm desirably retains geometric details and produces smooth surface.