DEVELOPING A LINE-OF-SIGHT BASED ALGORITHM FOR URBAN STREET NETWORK GENERALIZATION
Graphs composed of node and link are the most common way of representing street networks for any analysis beyond pure visual representation. An intuitive approach to convert a physical street network into an abstract mathematical graph is mapping street intersections as nodes while mapping street segments between intersections as links connecting nodes in the graph. Because this approach directly maps geographic entities into graph entities with the same dimension, i.e. zero-dimensional intersections to nodes and one-dimensional streets to links, it is called the primal approach.
Correspondingly, a dual approach is indirectly mapping street segments to nodes and intersections to links. Although the dual approach seems less intuitive, the dual representation of a street network it creates doesn't suffer from the same inherent low variance problem in the node's 'degree' as the primal representation does and therefore the dual representation usually exhibits the favorable scale-free and small-world properties of a network. Such characteristics of the dual representation makes it a better candidate for certain analyses where topological distance rather than metric distance plays a more important role. Space syntax, a well-known methodology for architecture and urban analyses, is one of the examples that rely on the dual approach.
The process of cartographic generalization is usually necessary before the creation of the dual representation because of street topology. Although there have been many cartographic generalization algorithms, few are tailored to satisfy the need of urban street network analyses based on the dual representation. This paper presents a generalization algorithm focusing specifically on urban street networks that utilizes the accompanying drawing of urban blocks and the concept of convex space, medial axis, and line-of-sight. The algorithm can then be implemented either as a modified v.generalize module with the addition of this new method or a new dedicated spatial network analysis module in GRASS.
Wen-Chieh (jeffrey) Wang - Chaoyang University of Technology