建筑物层次空间聚类方法研究

Hierarchical spatial clustering of buildings

建筑物空间聚类是实现居民地地图自动综合的有效方法。基于图论和Gestalt原理,发展了一种层次的建筑物聚类方法。该方法可以深层次地挖掘建筑物图形的视觉特性,将面状地物信息充分合理地表达在聚类结果中。依据视觉感知原理,借助Dealaunay三角网构建方法,分析了地图上建筑物的自身形状特性和相互间的邻接关系,并依据建筑物间的可视区域均值距离建立了加权邻近结构图,确定了建筑物的邻近关系(定性约束)。根据Gestalt准则将邻近性、方向性和几何特征等量化为旋转卡壳距离约束和几何相似度约束。通过实例验证了层次聚类方法得到更加符合人类认知的建筑物聚类结果。

Spatial clustering provides an effective approach for generalization of residential area in automated cartographic generalization.Based on graph theory and Gestalt principle, a hierarchical approach is proposed in this paper.This approach can be utilized to discover the graphical structure formed by buildings, which is obtained with the consideration of shape, size and neighboring relations.The neighboring relations are detern3ined by Delaunay triangulation, which is a qualitative constraint among buildings.A weighted neighboring structural graph is obtained by setting visual distance as the weight of the linking edge between adjacent buildings.Two levels of quantitative constraints are developed by considering the Gestalt factors,i.e.proximity, orientation and geometry of buildings.One is the rotating calipers minimum distance;the other is the geometric similarity measure.Through experiments it is illustrated that the results by the hierarchical spatial clustering proposed in this paper are consistent with human perception.