Major efforts to detect man-made symbols like house roofs and street lines have recently been made, but types of symbol such as single trees, tree groups or dry valleys are often neglected or of only minor interest.
A method based on shape descriptors has been applied to recognize single vectorised map symbols. Instead of trying to solve the map understanding process as a whole, the method focuses on the detection of specific symbols, resulting in efficient and operator-free interaction during the recognition process. Each symbol discriminator is defined by a set of rules based on the calculated shape descriptors. The definition process is the interactive part of the method, whereby the user fulfills the complex discrimination definition.
To detect whole tree groups or other aggregated symbols, a spatially sensitive grouping process has been chosen. The density function of small tree groups or dry valleys reveals a concentric or linear shape characteristic. A neighborhood-dependent triangulation method has therefore been developed to group the single map symbols. Starting with the nearest objects, triangulation „grows” until the single objects are too far away from other objects. The triangulation mesh of the identified objects yields the convex hull for the whole object. Further hierarchical triangulation with the gravitation centers of the base triangles yields generalized object clusters until a triangulation pyramid is formed, the so called Tetra-Tree. Similar to the Quad-Tree approach, a Tetra-Tree is only a temporary data structure used to fulfill the grouping process. In the case of top-level triangles of tree groups, a simple but elegant generalization of shape and direction is also generated.
To detect linear structures like dry valleys, the concentric growing behavior of the algorithm will be modified by a linearly growing triangulation process, with strongly elongated triangles being required to continue the triangulation.