T-spline has been recently developed to represent objects of arbitrary shapes in computeraided design, computer graphics, and reverse engineering. In fact, fitting a T-spline over a point cloud is usually ineffective by using traditional iterative fit-and-refine paradigm. In traditional T-spline least-square fitting method, all control points are recomputed in each iteration, which costs large amount of calculations. In this paper, we propose a fast T-spline fitting method based on T-mesh segmentation. The segmentation technology is introduced to identify the inactive and active region of T-mesh. Computational costs can be largely reduced since only the control points in the active part need to be recalculated in the upcoming process, while those in inactive part are kept invariant once the fitting accuracy is achieved. Classical datasets are used to validate the proposed fast fitting method, and the experimental results yield that a total running time is reduced to 34% of the traditional T-spline fitting method. We argue this method is particularly useful in the reconstruction of scanned scatter data of which the parameter distribution is not uniform.

Airborne light detection and ranging (LiDAR) technology is becoming the primary method for generating high-resolution digital terrain models (DTMs), which is essential for commercial and scientific uses. In order to generate DTMs, non-ground features as buildings, vehicles, and vegetation must be recognized and distinguished from the LiDAR point cloud. However, various degrees of errors may accumulate in the separated filtering and modeling processes. In this paper, a novel physical process driven DTM generating method for airborne LiDAR measurement is proposed, which combines the point cloud classification and surface fitting process simultaneously. Actually, the physical dynamic process is integrated with the common non-uniform rational b-splines (NURBS) model under the corresponding parameter mediation. The experimental results show that the proposed method is efficacious in reducing errors and have a nice performance in terrain fitting.INDEX TERMS Digital terrain model, physical process driven fitting, NURBS, LiDAR point cloud.

This paper proposes a coarse-to-fine registration algorithm for 3D point cloud. A novel feature points extraction method is presented, following an integrated local feature descriptor including Gaussian curvature, average curvature and point density for each point, through which we can achieve the coarse registration. Then, the ICP method is employed to refine the registration results with a good initial guess. Experiments including different simulated data sets demonstrate the applicability of the proposed methods. Meanwhile, the proposed coarse-to-fine registration algorithm is demonstrated to be robust to common nuisances, including noise and varying point cloud resolutions, and can achieve high accuracy and computation efficiency.