Automatic Detection of Patterns in Road Networks – Methods and Evaluation

Spatial data play a major role in many areas. They are an important source of information for a wide range of applications. The quality and quantity of data have increased in the last decades, resulting in a huge amount of data in the field of geoinformation. As a result of the multitude of information, various techniques for the automatic interpretation of large quantities of data have been investigated and implemented. Well known spatial data mining techniques, like spatial association rules, are primarily seeking for dependencies between attributes and spatial relations. In our opinion few investigations have been done for the detection of geometrical structures in vector data. In this paper we will concentrate on the topological and geometrical interpretation of line or route networks. Especially we will focus on the investigation of road networks, but applications and differences to railway and river networks will also be described. Such kind of vector data contains a great potential of knowledge, which is not given by means of explicitly stored geometric elements and their predicates but rather encoded as implicit knowledge in terms of topological connections, relations between elements, and typical structures or configurations of single geometric features. Typical road databases do not explicitly store such implicit information on patterns. Thus, a considerable need for interpretation of this data exists. First of all three basic patterns are introduced, namely grids, stars and ring roads. We will describe and analyse these patterns in detail and give algorithms to detect them automatically. Therefore we use an extended stroke approach, single source shortest path algorithm, theory of affine invariant moments and we perform a classification of a road network in urban and rural areas. We have developed an interactive control tool to realise these algorithms for pattern recognition and to display the results. Secondly, we will give an evaluation of these three patterns based on a comparison with a user test, in which test persons had to manually detect the described patterns in different test data sets and the test persons had to evaluate the patterns which were automatically detected with our approach. We will conclude the paper with an outlook to possible further tasks and applications of these patterns. The detected road network structures and their typical characteristics provide an opportunity for knowledge enhancement in domains like map generalisation, automatic map generation, visualisation of data, automatic annotation of databases, as well as information retrieval from the internet. * Corresponding author.