Abstracts Computer Science

GIS in a simulator environment and efficient inverse mapping of roads

by Stefan Langemark

This thesis investigates the possibilities of using GIS (Geographic Information System) data with an airborne autonomous vehicle developed in the WITAS project. Available for the thesis are high resolution (0.16 meter sample interval) aerial photographs over Stockholm, and vector data in a common GIS format containing all roads in the Stockholm area. A method for removing cars from aerial photographs is presented, using the filtering method normalized convolution, originally developed for filtering uncertain and incomplete data. By setting the certainty to zero over the cars, this data is disregarded in the filtering process, resulting in an image without cars. This method is further improved by choosing an anisotropic applicability function, resulting in a filtering that preserves structures oriented in certain directions. The available vector data is investigated with regard to its use in a simulator for vehicle movement, and is found to be missing much of the essential information needed in such a simulator. A new data format better suited to these requirements is created, using the extensible markup language (XML) which generates a humanreadable data format and can use existing parsers to make the implementation simpler. The result is a somewhat complex, but highly general data format that can accurately express almost any type of road and intersection. Cars can follow arbitrary paths in the road database and move with a smooth motion suitable for use as input to image processing equipment. The simulator does not allow any dynamic behaviour such as changing speeds, starting or stopping, or interaction between cars, takeovers or intelligent behavior in intersections. In the airborne vehicle, a mapping from pixels in a camera image (like the ones output from the simulator) to locations in the road database is needed. This is an inverse mapping with respect to visualizing as described above. This gives important information to a car tracking system regarding the probable movement of cars and also making it possible to determine if a car breaks traffic regulations. A mapping of this kind is created using a simplified form of ray tracing known as ray casting, together with space partitioning methods used to vastly improve efficiency. All above mentioned tasks are implemented using C++ and object oriented methods, giving maintainable and extendable code suiting a quickly changing research area. The interface to the simulator is designed to be compatible to the existing simulation software used in the WITAS project. Visualization is done through the OpenGL graphics library, providing realistic effects such as lighting and shading.