Approximating Bounded, Non-Orientable Surfaces from Points
Total Page:16
File Type:pdf, Size:1020Kb
Approximating Bounded, Non-orientable Surfaces from Points Anders Adamson Marc Alexa Department of Computer Science, Darmstadt University of Technology Fraunhoferstr. 5, 64283 Darmstadt [email protected] Abstract All approaches seem to assume that the represented sur- face is a solid and, thus, unbounded and globally orientable. We present an approach to surface approximation from This might stem from the fact that point clouds are mostly points that allows reconstructing surfaces with boundaries, acquired by scanning real-world (solid) objects, so that ar- including globally non-orientable surfaces. The surface is eas lacking points are considered inadequately sampled and defined implicitly using directions of weighted co-variances to be fixed. However, with the rising interest in unstruc- and weighted averages of the points. Specifically, a point tured point sets as a general surface representation, we feel belongs to the surface, if its direction to the weighted av- it is important to consider surfaces with boundary explic- erage has no component into the direction of smallest co- itly. Furthermore, enhancing reconstruction to faithfully variance. For bounded surfaces, we require in addition that deal with bounded, possibly non-orientable surfaces, in- any point on the surface is close to the weighted average of creases the robustness against sampling errors and makes the input points. We compare this definition to alternatives the algorithm more versatile. and discuss the details and parameter choices. Points on We extend our framework for approximating surfaces the surface can be determined by intersection computations. from scattered points [1] to handle smooth boundaries and We show that the computation is local and, therefore, no show that the surface could be (globally) non-orientable. globally consistent orientation of normals is needed. Conti- Specifically, we make the following contributions: nuity of the surfaces is not affected by the particular choice of local orientation. We demonstrate our approach by ren- Boundary definition: The surface is defined implicitly us- dering several bounded (and non-orientable) surfaces using ing compactly supported functions. We show how to ray casting. compute and define a smooth and natural boundary us- Keywords: point-sampled geometry, ray-surface inter- ing only the quantities we compute during surface ap- section, manifold, boundary, non-orientable proximation. Boundary behavior and parameters: We prove several properties about the behavior of the computations in- 1 Introduction volved in the estimation of the boundary. We use these approximations to solve the problem of parameter es- Point sets without additional topological information are timation and provide the user with an interval of rea- a popular representation of surfaces (see [3, 16, 32]). Real sonable parameters to choose from. world shapes are sampled using scanners resulting in an Orientation: The computation procedure for points re- initial surface representation consisting of points [21, 27]. quires only the intersection of a curve/ray with a plane. Points are also a suitable primitive for the display of sur- As the orientation of the plane’s normal is irrelevant, a faces, especially as the complexity of the shape results in global orientation of normals is not required. We show triangles being rendered into less than a pixel for the final that the continuity of the surface is not affected. display [26, 28, 18, 11, 10, 33]. For many modeling tasks, however, a (local) approxima- Despite these features, the point set could still define a tion of the surface based on the points is required. A variety solid and our implicit definition of the surface would allow of methods for the reconstruction of surfaces from points defining inside and outside. Note that we cannot handle exist. Some of these are approaches are global in nature, non-manifold surfaces, e.g., surfaces with self intersections. others allow locally computing surface parts but are based In the following, we briefly review some of the related work, on locally approximating a global implicit function. define the surface and explain how to compute it efficiently, Proceedings of the Shape Modeling International 2004 (SMI’04) 0-7695-2075-8/04 $20.00 © 2004 IEEE before we explain in detail the definition of the boundary are, e.g., positive. Our approach could, indeed, be inter- and how to choose the necessary parameters. preted in this way. 2 Related Work 3 Surface definition A variety of works tackle the (ill-posed) problem of re- We first give a brief definition of the surface as defined constructing a surface from an unstructured set of points. in [1] and establish several notations for later use. P = {p ∈ R3,i ∈{1,...,N}} Hoppe et al. [17] define a signed distance field from the Given points i on or S S points. For each point a normal direction is estimated and close to a surface . The point set surface P is defined as the normal is oriented. The signed distance to the surface the zero-set of the implicit function is defined as the normal component of the distance to the f(x)=n(x) · (x − a(x)), (1) closest point. x In a sense the surface definition of Hoppe et al. is related which describes the distance of to a plane with normal n(x) a(x) to Voronoi-based reconstruction techniques [8, 13, 4, 15, 6, through the weighted average of points .Forthe 5] as the surface is defined by a closest point relationship. surface definition to be reasonable (i.e. the surface is con- These techniques define the surface as a subset of the De- tinuously differentiable), we require the mappings defin- n : R3 → S2 launay triangulation of the point set. As such, they are all ing normal direction and weighted average 0 a : R3 → R3 global methods and generate a C surface. to be smooth in their input at least in a suf- Ω P Surfaces with higher order continuity from points are ficient neighborhood around . We will give particu- typically implicit. For the specific purpose of consolidat- lar choices for the mappings and the neighborhood below. ing registered range images Curless et al. [12] used a dis- These choices are useful for efficient and local computation crete representation (see also [23] for a comparable ap- of the surface, and they lead to natural definition of bound- proach). Radial basis functions with global [29, 9, 31] or aries. θ : R → R local [22, 19, 25] support are a more general technique for Let be a monotone decreasing function with r approximating functions from scattered constraints. Note local support θ, which is strictly positive inside the support, θ(χ)=0⇔ χ>r θ that all of these techniques are inherently global in nature i.e. θ. We will use as a weight function p x – some of them explicitly consider the case of incomplete determining the influence of a point i on a point in space . Ω point sampling and explain how, despite the missing points, The neighborhood is consequently defined as the region in space that is affected by the points: a globally unbounded and oriented surface would be recon- structed. Hierarchical methods based on blending individual local Ω= x| θ(||pi − x||) > 0 (2) implicit approximations such as [24] as well as moving least i squares type surface approximation [20, 3, 14] are local rep- Alternatively, we can define the distance of a point x to the resentations. While the MLS-based approaches explicitly point set as the minimum distance of x to any point in P: require the surface to be sampled sufficiently dense from a dP (x)=min||x − pi||. (3) closed manifold, hierarchical methods are rather concerned i with the problem of adequately filling un-sampled regions. Using this definition, Ω could be defined as the set of points Schaufler and Jensen [30] define the intersection of a ray with distance dP less than rθ, which also describes the and the point set as follows: In each point a disk is con- union of rθ-balls around the points in P. Note that all these structed using the point normal. A cylinder around the ray definitions of Ω are equivalent because we assume θ to be is intersected with the disks. The intersection is computed strictly positive in its support. as a weighted average of disks whose centers are inside the The weighted average of points a is defined (for points cylinder. Note that the reconstructed surface could have in Ω only) as boundary, because not every cylinder contains points. On the other hand, the boundary of the surface would depend piθ(||pi − x||) a(x)= i , x ∈ Ω. (4) on the particular rays chosen. i θ(||pi − x||) Our approach is also based on intersecting rays with the Normal directions are defined as directions of smallest surface [1], however, it could be understood as a robust weighted co-variance, using the same weights as before. computation of a particular implicit surface (which is iden- This could be understand as a least squares fit of a plane tical to the MLS definition [20]). Implicit surfaces typi- with unit normal n through x, i.e. the minimizer of cally represent solids. According to Bloomenthal et al. [7], 2 boundaries are generally represented by defining additional n, pi − x θ(||pi − x||) min i . (5) functions on the space and requiring that these functions ||n||=1 i θ(||pi − x||) 2 Proceedings of the Shape Modeling International 2004 (SMI’04) 0-7695-2075-8/04 $20.00 © 2004 IEEE It is well known that this constrained minimization prob- lem can be solved using the eigenvectors of the matrix W (x)={wjk} of weighted co-variances in the directions 3 of an orthonormal basis of R .Letei,i∈{0, 1, 2} be such a basis, then the coefficients of the matrix are given as wjk = ej, pi − xek, pi − xθ(||pi − x||).