期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2003
卷号：XXXIV-3/W8
出版社：Copernicus Publications
摘要：Road junctions are important objects for traffic related tasks, vehicle navigation systems, but also have a major role in topographic mapping. The paper describes an approach of automatic junction detection using raster and vector information: mean and standard deviation of gray values, edges as road borders etc. The derived feature set was used to train a feed-forward neural network, which was the base of the junction operator. The operator decides for a running window about having a road junction or not. The found junctions are marked in the output image. The operator was improved by additional features considering parallelism information of roads in junctions. The developed method was tested on black-and-white medium resolution orthoimages of rural areas. The results demonstrate the ability to find road junctions without preliminary road detection
