Object and feature recognition in a digital terrain model

A method for object and feature recognition is useful in many different areas. The most important application is perhaps simulations and other types of 3D-visualisations where there is a need for determining location and properties of a large number of different objects. Various terrain analysis problems could be solved such as drivability, where the detection of flat areas could be used to determine whether the terrain is suitable for driving vehicles, or possibly for landing-sites for helicopters. In this Master Thesis, a method for finding terrain-objects using a symbolic structure has been devised. The method is able to find long narrow, straight and winding ditches as well as ridges, hills and roads. This is accomplished by finding separate cross-sections unique to the terrain-object and then connecting these to complete objects. The cross-sections are found using patternmatching techniques. The evalutation of the algorithms has been made on a data-set from a laser-radar surveillance done with a TopEyeTM system.