The Extraction of Feature Lines on 3D Models: A Survey

Abstract

Feature lines are perceptually important features on 3D models, and have extensive applications such as mesh simplification, re-gridding, shape analysis, nonphotorealistic rendering and surface smoothing, etc.. This paper gives a comprehensive survey on this field, which has not been summarized before. According to whether feature lines are associated with the line of sight, feature lines can be classified into two categories: viewdependent curves and view-independent curves. Besides, view-independent curves can be further classified according to the type of 3D models, and view-dependent curves can be classified according to the property of feature lines. A complete classified method of feature lines on 3D models is proposed in this paper, and the challenges for feature extraction are also discussed briefly.

Feature line detection of noisy triangulated CSGbased objects using deep learning

“…Several detection methods summarized in [4] identify feature lines using local surface curvature and normal estimation. For discrete surfaces, there is only an approximation of the local curvature and vertex normal vectors.…”

“…For discrete surfaces, there is only an approximation of the local curvature and vertex normal vectors. Therefore, there are several different approaches calculating the local curvature, such as averaging through the neighbourhood of the face or interpolating a local polynomial patches [4]. In particular [11] uses a mesh curvature histogram to determine sharp and smooth edges.…”

“…In several recent publications, machine learning was used for feature line detection [2] [9] [10] [1], which can divide the 3D objects into meaningful surface patches. Since feature lines can be determined locally [4], a local detection by selecting a local patch similar to [10] [9] [1] is basis for our approach. In particular, [10] uses a feature line detection of sharp edges based on point clouds using a deep learning model trained on meshed data.…”

“…Moreover, manual reconstruction itself can lead to inaccurate approximation of the surfaces by human preparation and interpretation, especially for noisy and deformed triangular meshes. As a solution, the robust detection of important lines called feature line will allow an automatic reconstruction of such discrete surfaces [4]. Then a proper reconstruction of the mesh allows the CAD geometry to be updated by using the surface of a deformed finite element mesh, that has been analyzed [5] [6] [7].…”