AHD: The alternate hierarchical decomposition of nonconvex polytopes (generalization of a convex polytope based spatial data model

Please download to get full document.

View again

of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Share
Description
Robust convex decomposition, RCD, of polytopes is the convex decomposition of nonconvex polytopes using algorithms whose implementation is based on arbitrary precision arithmetic. Decomposing nonconvex polytopes using RCD can make the data
Transcript
  AHD: The Alternate Hierarchical Decomposition of  Nonconvex Polytopes (Generalization of a Convex Polytope Based Spatial Data Model) Rizwan Bulbul Institute of Geoinformation and Cartography Technical University of Vienna Gusshausstrasse 27-29 E127 A-1040 Vienna, Austria  bulbul@geoinfo.tuwien.ac.at Andrew U. Frank Institute of Geoinformation and Cartography Technical University of Vienna Vienna, Austria frank@geoinfo.tuwien.ac.at   Abstract   —Robust convex decomposition, RCD, of polytopes is the convex decomposition of nonconvex polytopes using algorithms whose implementation is based on arbitrary precision arithmetic. Decomposing nonconvex polytopes using RCD can make the data representation model consistent enabling generalization with level of detail. Our approach, alternate hierarchical decomposition, AHD, for the decomposition of nonconvex polytopes with arbitrary genus, is a recursive approach whose implementation is robust, efficient and scalable to any dimension. Our approach decomposes the given nonconvex polytope with arbitrary genus into a set of component convex hulls, which are represented hierarchically in a tree structure, convex hull tree, CHT.  Keywords-convex decomposition; hierarchial; polygon; convex hull tree I.   I  NTRODUCTION (REPRESENTATION   OF   GEOMETRY   OF   REGIONS) An irregular subdivision of space contains nonconvex regions. A region is defined as a set of polygons [1], most of them may be nonconvex. The representation of geometry of (convex) regions as a collection of inequalities as y ≥  ax + b has been proposed in [2, 3]. Our approach for the representation of geometry of nonconvex regions is based on the representation of (convex) polytopes, which are defined by the intersection of finite set of half planes (half spaces in case of 3D). We use the representation of polytopes in dual space, which are represented by set of points and lines, depicting lines and the connection points of half planes in the Euclidean space respectively. In dual space, a convex polytope is defined as the convex hull of the dual points [4]. Our model for the representation of region is starting with the convex hull and removing convex parts to describe concavities (e.g. union of convex polytopes). The object representations based on simpler geometric structures are more easily handled [5] than complex structures. The algorithms for convex objects are simple and run faster than for arbitrary objects [6-9], for example, the decomposition of complex spatial objects into simple components considerably increases the query processing efficiency [6]. Therefore, the need arises to decompose nonconvex polytopes into simpler convex components. The problem of decomposing a nonconvex object into its convex components is known as “convex decomposition” and has applications in diverse domains ranging from pattern recognition, motion planning and robotics, computer graphics and image processing, etc. [10-14]. The application of decomposition techniques in the spatial data modeling domain is sparse in literature, only few address the issue in the spatial database domain [6, 15]. There is a need of researching the decomposition techniques in the spatial data modeling domain, thus developing robust, efficient, and practical algorithms for the decomposition of complex spatial objects into simple components. Our convex decomposition approach is simple in which the srcinal input, nonconvex polytope with or without holes, is decomposed into its convex components that are convex hulls arranged hierarchically in a tree structure (details in section 3). Most of the decomposition solutions are based on the real RAM (Random Access Machine) model of computation. Our approach is robust as the underlying algorithms for recursively computing convex decompositions have been implemented using big numbers, which provide arbitrary precision. RCD of  polytopes is the convex decomposition of nonconvex polytopes using algorithms whose implementation is based on arbitrary  precision arithmetic. Decomposing nonconvex polytopes using RCD can make the data representation model consistent enabling generalization with level of detail. Our basic strategy is similar to the approach in [16] and the sum-difference approach [9, 17]. We use the QuickHull   algorithm [18] for computation of convex hulls and all the procedures have been implemented as Haskell [19] modules. The solution is compact, efficient and robust because of the Haskell's built in list comprehensions, lazy evaluation, and support for big integers. II.   CLASSIFICATION   OF   APPROACHES   TO   POLYGON   DECOMPOSITION Polygon decomposition approaches are classified based on the following criteria [6, 10, 17]; Authorized licensed use limited to: Universitatsbibliothek der TU Wien. Downloaded on November 12, 2009 at 08:13 from IEEE Xplore. Restrictions apply.   A.   Type of subpolygons (components) Depending on application, different decompositions can result in variety of components, e.g, convex, star-shaped, or fixed orthogonal shapes.  B.   Type of polygon being decomposed Decompositions can also be classified on the basis of input  polygons, e.g., closed or open, simple or connected, with or without holes, etc. C.    Relation of subpolygons Decompositions that result in components, which do not overlap except at boundaries, are classified as partitioning. Other decompositions allowing component overlaps are classified as covering.  D.   Objective function Most of the decompositions are classified based on the objective function. The two broad categories are decompositions that minimize the number of convex components and decompositions that minimize computational complexity in terms of execution time.  E.    Dimension specific or dimension independent Decomposition approaches can also be classified depending on the input polygon dimension. The majority of the decomposition techniques in literature are dimension dependent and mostly applicable in 2-D and 3-D [12] cases only. The spatial data representation model in [15], uses hierarchical convex based polygon decomposition approach like our approach for multi scale organization of vector data on spatial data servers. Their approach has a run time of O (n log  2  n) a   and is complex separately dealing holes from the rest of the  polygon and no implementation details provided. The work of [17] focuses on decomposition of orthogonal polygons. Another hierarchical strategy in [10] decomposes a polygon with or without holes into its approximate convex components. Their algorithm results smaller set of approximately convex components efficiently in O (nr) b . In [12] a triangulation based algorithm for the problem of partitioning a polytope in  R 3 has  been proposed. The algorithm requires O (n+r  2  )  space and runs in O ((n+r  2  )   log r)  time. A similar work in the pattern recognition domain has been done in [20], presenting a convex hull based algorithm for concavity tree extraction from pixel grid of 2-D shapes. Most of the algorithms for decomposition are optimized for one of the criteria [6], depending on application needs. Table 1 shows a characterization of our approach based on the aforementioned criteria. a   n is the number of vertices    b   r is the number of notches/delta regions/concavities   TABLE I. C HARACTERIZATION OF OUR DECOMPOSITION TECHNIQUE   Criteria Our Approach Type of components Convex Type of input Polytope with or without holes Relation of components Covering- as the component convex hulls at level l+1  are contained within the convex hull at level l   Objective function Does not fall to any category, but our objective is to achieve generalization of data modeling framework using convex polytopes. Input dimension Dimension independent III.   OUR APPROCH ( ALTERNATE HIERACHIAL DECOMPOSITION ) Our approach, AHD  ,  for the decomposition of nonconvex  polytopes is a recursive approach whose implementation is robust, efficient and scalable to any dimensions. Our approach decomposes the given nonconvex polytope into a set of component convex hulls that are represented hierarchically in a CHT   (related concepts are convex deficiency tree  [18] and concavity tree  [20]). CHT   is a tree of convex hulls in which the root node contains the convex hull of the input polytope, non-leaf nodes contain convex hulls of the nonconvex delta regions (notches) of the nonconvex structure whose convex hull is represented by parent node. Leaf nodes contain the convex hulls of the convex delta regions. The convex hulls in CHT   at different levels are assigned positive (+) and negative (-) signs alternately as shown in Fig. 2. The convex hulls at even and odd levels have positive and negative signs receptively. The  positively signed convex hulls represent the convex regions to  be included while the negatively signed convex hulls represent convex regions to be excluded. The AHD consists of two steps: (1) convex hull computation, and (2) delta region/s extraction (using symmetric difference). The process is then recursively applied to the delta regions until the input region is convex. The necessary invariants for our approach are; •   The input is a (nonconvex) polytope with or without holes represented in the dual space, which is a closed region of edges. •   An edge is a pair of points and each point has homogeneous big integer coordinates. •   The boundary edges are oriented in counter clockwise direction while the hole edges are oriented clockwise. •   In CHT  , the convex hulls at level l+1  are inside their respective parent at level l.  In order to explain our approach, we present here an example that demonstrates the decomposition steps and shows the output CHT. Suppose that we are given a 2-D nonconvex  polytope with a hole (H) in dual space (as shown in Fig. 1a) as input. The application of AHD   results in the decomposition of the input into its component convex hulls (Fig. 1b). The complete AHD process is illustrated in Fig. 6. Authorized licensed use limited to: Universitatsbibliothek der TU Wien. Downloaded on November 12, 2009 at 08:13 from IEEE Xplore. Restrictions apply.    (a)   (b)   Figure 1. Input polytope (a) and its decomposition (b) The output is a hierarchical representation of convex hulls as CHT as shown in Fig. 2. The data structure used for CHT representation is an arbitrary tree, every node of which contains a convex hull. We have implemented CHT in Haskell using  Rose tree  [21], which recursively defines an arbitrary tree as: Tree = Node  CHull [Tree]. Thus every node contains a convex hull and may contain one or more children trees represented as a list of Tree with the associated node. The tree that has no children is leaf node defined as leafNode =  Node  CHull [ ]. IV.   THE   ALGORITHM   IN   PSEUDOCODE Our AHD process has the function “build”, which decomposes the given input polytope into its convex hull components and the component hulls are hierarchically arranged into the CHT. The pseudocode for “build” function is as given in Fig. 3. The pseudocode of the algorithm for computing the QuickHull [18] is given in Fig. 4. Figure 2. CHT for the demonstration example of Fig. 1 Figure 3. Pseudocode of the algorithm ‘build’ that populates the CHT (a) (b) Figure 4. Pseudocode of the algorithm quickhull (a) and its helping function qHull, which computes the sub hulls of partitions recursivley Once the decomposition is done, the resulting CHT can be  processed to retrieve the decomposed object from its convex hull components at any level of detail depending on application. The pseudocode of the algorithm for complete  processing of a CHT, which retrieves the srcinal object, is given in Fig. 5. H Authorized licensed use limited to: Universitatsbibliothek der TU Wien. Downloaded on November 12, 2009 at 08:13 from IEEE Xplore. Restrictions apply.  Figure 5. Pseudocode of the algorithm ‘process’ that processes the CHT to extract represented region V.   IMPLEMENTATION   IN   HASKELL The algorithms have been implemented as Haskell modules. The selection of Haskell provides ease of implementation giving simple and compact code. This is  possible because of Haskell's built in “list” data structure and its associated higher order functions, lazy evaluation and support for big numbers. Most computational algorithms are designed with the assumption that all the numerical computations involved give exact results. In reality, the results of numerical computations in current systems are not exact. Results are rounded to the nearest approximates, and inconsistencies can occur because of the round-off errors, the problem called nonrobustness. Thus the accuracy of numerical computations in current systems is limited by the finite precision arithmetic. The problem is more severe in case of geometric algorithms [22], as these involve  both metric and topological processing. The rounding errors in metric computations because of the fixed precision arithmetic can cause inconsistencies in topological processing, leading to inaccurate geometric computation models. In the polygon decomposition domain, the issue of inaccuracies in numerical computations due to fixed precision arithmetic is considered in [9, 20]. To achieve a robust implementation of AHD, we are using the homogeneous big integers for coordinate representation of half planes in dual space. The viability of using big integers for metric processing has been experimented in [23]. It has been found that contrary to the widespread belief, programming with  big integers is straightforward and easier than using conventional integer or floating point data types. Metric computations with big numbers are conceptually simpler and need no testing for approximation problems. VI.   SOLUTION   CHARACTERISTICS  A.    Robustness The implementation of our approach avoids the problem of numerical nonrobustness [24-27] by using exact algebra. This has been achieved through use of big integers allowing arbitrary precision arithmetic. Thus, topological inconsistencies are avoided, as the underlying algorithms have been implemented keeping in mind the arbitrary precision arithmetic for numerical computations involved.  B.    Level of Detail Depending on the applications, the complete decomposition of nonconvex polygons to convex polygons may not be needed [10]. In such cases approximations are used by halting the decomposition process using some threshold. Our approach is flexible in that the decomposition process can be stopped at any level depending on application thus allowing level of detail approximation. C.    Dimension independence To the best of our knowledge, the algorithms for decomposition problem are not dimension independent as most work for 2-D and some deal with 3-D objects. Although for demonstration purposes we are using 2-D polytopes, our solution is easily extendable to n-dimensional polytopes. This is achievable if the procedure for computing convex hull is dimension independent. For this any of the existing dimension independent convex hull algorithms [28, 29] can be used.  D.    Number of Components The solutions to the problem of decomposing a nonconvex  polygon into its optimal (minimum) number of convex components vary depending on whether Steiner points allowed [10]. Although our solution was not aiming to optimize the number of resultant components, however we found that our solution is not as bad as the mostly used decompositions for near optimal number of components. The number of components is the number of CHT nodes, which is 13 and  partitioning the example in Fig. 1 results in 14 convex components.  E.   Complexity Analysis Given a nonconvex polytope  p  with n  points, the convex hull and delta regions of  p  can be computed in O (n log n)  and in liner time, respectively. If r   is the number of delta regions of  p  then complexity becomes O (r × n log n). However, size of r is negligibly smaller than n  for higher values of n . Thus, the worst case scenario can not be O (n 2 log n)  and our approach has an average time complexity of O (n log n).  VII.   SUMMARY   AND   FUTURE   WORK The implementation of our decomposition approach is efficient, robust, and simple. A dimension independent implementation of our approach has already been implemented using the incremental convex hull algorithm [28] and the work is currently in writing phase. In addition, the CHT has been tested to support basic Boolean operations of intersection, union and symmetric difference. The results are in draft to be  presented soon. In future, we will test the support for operations for dimension independent implementation of CHT and experiment with real data sets. Authorized licensed use limited to: Universitatsbibliothek der TU Wien. Downloaded on November 12, 2009 at 08:13 from IEEE Xplore. Restrictions apply.    Figure 6. The AHD process steps. L represents the level, dr represents the delta region and ch represents the convex hull Authorized licensed use limited to: Universitatsbibliothek der TU Wien. Downloaded on November 12, 2009 at 08:13 from IEEE Xplore. Restrictions apply.
Advertisement
Related Documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks