Detecting road junctions by artificial neural networks

Road junctions are important objects for all traffic related tasks, and are essential e.g. for vehicle navigation systems. They also play a major role in topographic mapping. For automatically capturing road junctions from images models are needed, which describe the main aspects. This paper presents an approach to road junction detection based on raster and vector information. The raster features are similar to the ones used in classification approaches. The vector features are derived from a road junction vector model containing edges as road boarders. The whole feature serves as input to an artificial neural network. The neural classifier decides for search window, whether its central pixel is a part of a road junction or not. The developed junction operator was tested on several black-and-white medium resolution orthoimages. The achieved results demonstrate that such junction models can successfully identify three- and four-arms road junctions.