Surface modeling using multi-view range and color images

Accurate integration of multi-view range images using k-means clustering

“…• 2D mesh-based integration method [18][19][20] usually first requires triangulation of each range image into a mesh, and then detection of the overlapping area between the triangulated meshes from neighbouring views. The overlapping triangles are often detected using overlap polytopes [21], projecting triangles to image plane to see if they overlap and their distance is smaller than a threshold [19], or point to triangle distance [18].…”

“…The overlapping triangles are often detected using overlap polytopes [21], projecting triangles to image plane to see if they overlap and their distance is smaller than a threshold [19], or point to triangle distance [18]. The overlapping triangles are often processed in three ways: (1) the most accurate triangles in the overlapping area are kept and inaccurate ones are removed, all remaining triangles are then reconnected using some kinds of criteria such as maximizing the interior angles of triangles [19]; (2) the overlapping boundary is identified and the meshes on the two sides of the boundary from different images are then zippered together [18]; and (3) the boundary points and those in the overlapping area are re-triangulated [21]. Although these kinds of algorithms can overcome some shortcomings of the 3D volumetric integration algorithms above, they still have some drawbacks such as inaccurate detection of overlapping areas due to fixed distance thresholds, computational expensiveness due to solving the intersection points between overlapping triangles, and non-smoothness of the final surface.…”

“…[24], considering the confidence of their measurement. This kind of method is computationally efficient, since all processing is based only on points whose number is significantly smaller than that of the triangles used as primitives in the 2D mesh-based integration method [18][19][20]. The final surface tends to be rough due to imaging noise and registration errors.…”

An Evaluation Method for Multiview Surface Reconstruction Algorithms

“…The similar phenomenon also lies in the Frog model where the result of the quantitative evaluation via integration error was inconsistent with the qualitative comparison. [3], mesh-based method [15], SFK method [20], k-means clustering-based method [19], pairwise MRF-based method [11], higher-order CRF-based method [14] 1.2. Requirements for a good measure An evaluation method for multiview surface reconstruction is a specific measure of shape dissimilarity.…”

Scan integration as a labelling problem

“…Mesh-based methods [17,18,19] detect overlapping regions between triangular meshes. Then, the most accurate triangles in the overlapping regions are kept, and all remaining triangles are reconnected.…”