Abstract

In this paper, we present a new deformable model 2.5D Active Contour that is capable of directly extracting shape geometry from 3D unorganized point cloud datasets. The reconstructed surfaces are either open or closed. Furthermore, the new model can reconstruct and discover non-manifold geometry hidden in the point-cloud dataset. Our algorithm starts from a simple seed (e.g. a triangle) that can be automatically initialized, and always enlarges the model boundary outwards along its tangent direction suggested by the underlying dataset. This mechanism ensures that our novel models flow directly over the data boundary through the expansion of its bounding contour. To maintain the regularity of the model and the stability of the numerical integration process, commonly used mesh optimization techniques are employed throughout the deformation process. In addition, the new model can recover fine details of the underlying shape through local/adaptive refinements. We demonstrate both the accuracy and the robustness of our algorithm through a number of experiments on both real and synthetic datasets.