Research On Buildings Clustering Methods Based On Multi-Constraints

Achieving automated map generalization is a final goal of the multi-scale map data conversion, and this conversion process must follow the relevant mapping constraints. Taking the residential automated map generalization as example, the paper analyses the Gestalt constraints for a group buildings on the residential map. Main tasks are described as follows:1. Firstly we introduced the background of the buildings clustering research, and the residential cartographic generalization concepts, then reviewed the research situation, and put forward the problems of existing methods.2. Based on the hierarchical clustering method, the constraints for buildings clustering had divided into global and local constraints. There were two stage processes:firstly, based on the global constraints (the urban morphology method) process; secondly, the local constraints (the Gestalt principle) clustering process. In this section, the studies focused on the local constraints for building a single level clustering graph TIN method, and then get the MST proximity graph. Finally, the importance of the proximity, orientation and similarity criteria for buildings clustering was evaluated.3. Researching the distance constraints for proximity criteria, and indicating the existing methods for concave buildings, the RCAveg-Dis method was proposed. The method first decomposed buildings then clustered them, so the features vertex information was constrained in the process. Finally, based on the actual data test, clustering results showed that the RCAveg-Dis method can be more accurately get the buildings clusters.4. On the direction and similarity criteria, firstly, constraint methods for building's single direction and single shape were introduced. secondly, a number of buildings'geometric characteristics calculation methods were given. Finally, based on integration geometric constraint, the similarity constraint clustering method was proposed, and a large number of experiment results were illustrated. On the next buildings clustering level, the actual data verified the validity of the method.Finally, this paper makes a summary and presents some issues for future works.