Visual tools to select a layout for an adapted living area Sébastien AUPETIT, Arnaud PURET, Pierre GAUCHER, Nicolas MONMARCHÉ and Mohamed SLIMANE Université François Rabelais Tours, Laboratoire d Informatique, 64 avenue Jean Portalis, 37200 Tours, France {aupetit,arnaud.puret,gaucher,monmarche,slimane}@univ-tours.fr Abstract. In this paper, we present several visual tools used in the HM2PH project in order to include the future occupant in the design of her/his adapted house. The common goal is help the occupant to choose between a set of topological layouts. Each tool presents different characteristics and properties. Keywords. visualization, adapted living area, conception 1. Introduction and context In this paper, we present a part of our work on the HM2PH project (Habitat Modulaire et Mobile pour Personnes Handicapées 1 ) [6,5,11] which is developed at the Laboratoire d Informatique de l Université François-Rabelais de Tours. The main HM2PH goal consists in designing a affordable mobile and adapted living area for disabled persons. To simplify the design process, we rely on softwares. As other studies (see [3]), we need to take into account the specification of the specialists but we also aim at the inclusion of the future occupant into the design process. With such aim in mind, we split the design of the house into several steps (see Figure 1) grouped into three parts: the definition of the required features of the house, the layout design, and the layout validation and finalization. The first part can be half-automated. Laws, recommendations and information databases are automatically used but assistive technologies and wishes of the person are defined by the specialist and the future occupant. They define all the characteristics that the living area must satisfy. The second part is concerned with the design of a layout that satisfies the needs. The third part aims to validate and to finalize the selected layout. If the validation fails, a feedback process can be used to backtrack to the layout design part. The layout design part is organized according to the method we have used to automatically compute valid layouts from the needs. There is a lot of methods to compute satisfying layouts such as algebraic methods [2], deductive methods [9], methods with local propagation [4] or Constraint-Satisfaction-Problem based methods [12]. In our work, we considered a Constraint-Satisfaction-Problem based method [7,8,11] for many rea- 1 Modular and mobile house for disabled persons
Figure 1. The steps of the HM2PH design process sons (see [11] for a full explanation). Let us recall the main ones. The computed layouts fully respect the explained constraints. The layout generation can be split into two processes [7,8]: a generation of topological layouts and a geometric optimization. This split has for advantage to greatly reduce the computing cost. A topological layout is a global organization of rooms in the house that can be considered as an abstract layout or a draft layout. For a topological layout, dimensions, orientations and positions of the rooms are defined by ranges of values. All layouts, that can be derived from it, share the same organization and satisfy the needs in a same manner. Moreover, an exhaustive generation of topological layouts is tractable. To obtain a concrete layout for which each numerical ranges has only one value, a geometric optimization process is applied. It consists in reducing each range of values of a topological layout to the best suited value for a given criterion while satisfying the needs. The obtained layout is then a concrete layout. When a single concrete layout has been chosen by the occupant, it is needed to layout the required equipments and furnitures in the house to provide a realistic model of the house and to be able to validate the layout in a concrete realistic manner. During the overall process, the future occupant can decide if a topological layout or if the inteior is well organized or not. Automatic and visual validation tools [11] of the chosen layout are provided to help in this task. This validation and the finalization of the layout form the last steps of the design process. From the needs, lots of topological layouts can be generated. Moreover, the geometric optimization of a topological layout is time consuming. It is therefore impossible to apply the geometric optimization to all the topological layouts. But, very few topological layouts are usually interesting for the future occupant. Then, we need to choose few suitable topological layouts on which the geometric optimization is performed. A way to achieve it is to provide various visual tools to select layouts to the future occupant. Such tools are presented in this paper. 2. Key principles to compare the properties of topological layouts To compare topological layouts, we propose to follow two complementary approaches: a continuous one and a discrete one. The continuous approach considers pairwise dissimilarities between layouts. This dissimilarity is defined by the future occupant from interesting characteristic rooms such as the corridors or the kitchen. It evaluates how those
The black level shows the probability that there is a part of a corridor at this location. The first three representations are related to virtual layouts and the last to a concrete layout. Figure 2. Representations from four topological layouts showing where the corridors are located. (a) Positionning of the clusters (b) Inspection of a cluster Figure 3. Two-stages radial representation of the layouts. rooms are positioned and sized equally in two topological layouts (see [11] for details). It has been formally shown (in [10,11]) that this dissimilarity is a distance and that an euclidean embedding of the topological layouts can be built. Such embedding allows to consider a layout as a vector/point. It allows to manipulate a vector space of both concrete and virtual topological layouts. The virtual topological layouts can be combined for example as a weighted barycenter. The intrinsic properties of the dissimilarity allows to build a graphical representation from a virtual topological layout displayed with colors (showing where the interesting characteristic rooms are located on the layout) (see Figure 2). The second approach, which is a discrete one, considers the presence or absence of interesting rooms on nine locations/sectors (North, North-Est, Est, South-Est, South, South-West, West, North-West, Center). A topological layout is then ranked as verifying or not the presence of a type of rooms in a particular location. These two approaches lead us to define three kinds of visual tools to select a topological layout. 3. Selection of topological layouts by the future occupant 3.1. Two-stages radial representation The two-stages radial representation considers the dissimilarity approach (see [1,11] for details). It consists in clustering topological layouts with the K-means method and to represent the center of the clusters on a circle. The centers are positioned in order that the angle between two classes is representative of the dissimilarity between them and such that the sum of the distances between all the classes following the order on the circle is minimum. Virtual layout representations are used to show the center of the clusters and to show the average of all the topological layouts. The user can interactively inspect the content of a cluster and see the topological layouts composing the cluster by clicking on the center of a cluster. Figure 3 gives an example.
Figure 4. Nicheworks representation of a HAC built from the topological layouts. Figure 5. The interactive filtering method. 3.2. Hierarchical Ascendant Classification and the Nicheworks representation The second visual tool is based on two well known techniques: the Hierarchical Ascendant Classification (HAC) and the Nicheworks representation (see [10] for details). The pairwise dissimilarities are considered and a HAC is built on the representatives of the layouts. The hierarchy is plotted using the Nicheworks representation [13] which plots a tree using a concentrical approach and using an angles splitting process. Each sub-tree is plotted recursively on a screen space proportional to its number of leafs. In our tool, the leaves correspond to the topological layouts and the inner nodes correspond to the virtual topological layouts. Using the previously described representation of virtual topological layouts, the tool allows to progressively choose a layout (on the leaves) from the mean of all layouts (the root of the tree). Figure 4 gives an example. 3.3. Interactive filtering The third visual tool, which is also the most intuitive, considers the presence or the absence of interesting rooms on given location. The tool can be qualified as a filtering method. A filter rejects all layouts for which the interesting characteristic (type of rooms and location) is not present. For each filter, the list of valid topological layouts is displayed independently of the other filters. Then a last list is displayed showing only the layouts passing simultaneously all the filters. The user can interactively remove, add or change a filter. Figure 5 gives an example. 3.4. Differences between the three visual tools The three tools have been designed with different aims and show different properties: the aim (what can we do with it?), the intuitiveness of the analysis (do we need knowledge to understand the representation?), the simplicity (do we need to set up lots of parameters to use it?) and the abstraction capability (does the method allow to obtain an abstract view of the data?). The Table 1 gives the properties of the tools.
Table 1. Properties of the three visual tools. High rating is the best. Two-stages radial representation HAC combined with NicheWorks Interactive filtering Aim Trends search Progressive choice Filter layouts Intuitiveness of the analysis low medium high Simplicity low medium medium Abstraction capability medium high low 4. Conclusion In this paper, we have presented three visual tools to include the future occupant in the process of designing the future adapted house. The three tools are using different key principles and methods leading to different properties. The two-stages radial representation is more suited as an initial trend search in the set of topological layouts whereas others (HAC combined with NicheWorks and Interactive filtering) are more suited to make a decision by progressive choices or by filtering the layouts. In the future, we plan to link the tools together allowing to switch from one to another while keeping the context (the current cluster, the current layout... ). References [1] S. Aupetit, A. Puret, P. Gaucher, N. Monmarché, and M. Slimane. Classification automatique et visualisation de plans d habitations adaptées. Handicap 2006, pages 81 88, 2006. [2] Philippe Charman. Gestion des contraintes géométriques pour l aide à l aménagement spatial. PhD thesis, INRIA, 1995. [3] M. Ferreira, J. M., T. Amaral, D. Santos, A. Agiannidis, and M. Edge. The custodian tool: Simple design of home automation systems for people with special needs. EIB Scientific Conference, 2000. [4] G. Kwaiter. Etude et développement d un modeleur déclaratif 3D temps réel d environnements virtuels, basés sur les métaheuristiques issues de la recherche locale. PhD thesis, Université Paul Sabatier de Toulouse, 1998. [5] J. Leloup. Spécification d un espace vide mobile et adapté pour des personnes en déficit d autonomie. PhD thesis, Université François Rabelais, Tours, 2004. [6] J. Leloup, P. Gaucherand J. Garcia, and J. Siffert. The hmph project: Software for the design of an adapted living area. AAATE 2003, pages 559 563, 2003. [7] B. Medjdoub and B. Yannoub. Separating topology and geometry in space planning. Computer-Aided Design, 32(1):39 61, 2000. [8] B. Medjdoub and B. Yannoub. Dynamic space ordering at a topological level in space planning. AI in Engineering, 15(1):47 60, 2001. [9] D. Plemenos. Contribution à l étude et au développement des techniques de modélisation, génération et visualisation de scènes. PhD thesis, Université de Nantes, 1991. [10] A. Puret, S. Aupetit, P. Gaucher, N. Monmarché, and M. Slimane. Selection by visualization of topological layouts for adapted living area design. LNCS : ICCHP 2006, 4061:500 507, 2006. [11] Arnaud Puret. Projet HM2PH, Génération automatique de plans et visite virtuelle d habitats adaptés pour personnes handicapées. Thèse de doctorat, Université François Rabelais de Tours, 2007. [12] O. Le Roux, V. Gaildrat, and R. Caubet. Design of a new constraints solver for 3d declarative modelling. 3IA 2000, pages 75 87, 2000. [13] G. J. Wills. Nicheworks - interactive visualization of very large graphs. Journal of Computational and Graphical Statistics, 8(2):190 212, 1999.