Partenaires et organisations internationales
(Anglais)
|
Institut Géographique National (F) (IGN, Project Leader), Inst. Nat. Polytechnique de Grenoble, MAGMA (F) (INPG), University of Edinburgh (UK) (UNI-ED), Laser-Scan Ltd., Cambridge (UK) (LSL), Universität Zürich, Geographisches Institut (UNI-ZH)
|
Résumé des résultats (Abstract)
(Anglais)
|
In response to industrial needs, the project aimed to develop expertise, techniques and a research prototype for automated map generalization (i.e., the derivation of maps and spatial data products at reduced scales) as part of the functionality of Geographical Information Systems (GIS). The project brought together three technical domains, represented by the expertise of the five project partners: automated map generalization; object-oriented and active spatial databases; and multi-agent systems (MAS).
One of the main problems in automating map generalization is the 'orchestration' of algorithms, given multiple and most often conflicting constraints. The AGENT project introduced a new model of map generalization based on the notions of autonomy and constraints. These ideas were implemented using a MAS approach, whereby cartographic constraints represent goals of the agents. Agents are recursively formed from map objects: micro-agents represent single cartographic objects (e.g. a building or a road) while meso-agents represent groups of objects (e.g. city blocks or road networks). Micro-agents operate under the control of meso-agents. Each agent makes use of measures (methods to detect spatial conflicts and evaluate the satisfaction of its goals) as well as generalization algorithms (methods to resolve spatial conflicts by geometric transformations). The application of the measures and generalization algorithms is controlled by the constraints that apply to a particular type of agent (e.g. a building should be large enough, wide enough, shouldn't change its position too much, etc.). Thus, each agent has a variety of relevant measures and generalization algorithms associated. A hill climbing algorithm is used to find the solution best satisfying the constraints (goals) of an agent. AGENT most probably represents the most thorough use of MAS technology in the GIS domain.
A prototype system was developed based on Laser-Scan's LAMPS2 object-oriented GIS platform. Apart from the MAS engine, various new algorithms for map generalization as well as for spatial measures were developed and implemented. The functionality of the prototype is focuses on the generalization of road networks as well as urban and semi-urban settlements but available methods could be modified to deal with similar object classes. The basic architecture of the prototype is easily extensible, as experience with data and requirements from mapping agencies not represented in the project has shown.
Immediately following the end of the project, the commercial partner, Laser-Scan Ltd., has started the development of the AGENT prototype into a commercial software product and has been able to attract initial customers among leading national mapping agencies. Also, one of the project partners (IGN France) is using this software for its Carto2001 project (automated production of the French 1:100,000 map series).
|
