Automated extraction of road networks from satellite images for preparing and updating road location data for geographic information systems in transportation (GIS-t)

Transportation agencies apply GIS (Geographic Information System) technology to better manage spatially distributed transportation facilities and services. However, due to high data conversion and update costs, GIS technology can be expensive. Remotely sensed images are widely recognized as a ready data source that might lower the cost of CIS considerably. One data object critical to transportation applications of GIS that can be extracted from remotely sensed images is the road network. This research presents a three-level (low, intermediate, and high) automated process capable of extracting road network location attributes from SPOT panchromatic images. Low-level operations extract road pixels from gray images. Intermediate-level operations convert road pixels into vectorized links which are usually fragmented. High-level operations defragment links via line-linking functions to construct road networks. This research focuses on improving the various existing techniques employed in each operation level. A comprehensive computer program was developed to implement all algorithms.;For low-level operations, extensive testing was conducted to find a suitable line detector; an automated threshold selection strategy for the line detector was developed; a thinning algorithm was adapted to reduce lines to one pixel in width; and a noise removal algorithm was developed to remove pixels that form either short unconnected lines or strokes.;Improvements to intermediate-level operations include developments of an adaptive binary image decomposition structure for the Hough transform, a heuristic algorithm to suppress redundant lines in the Hough output, and an adaptive thresholding scheme for the Hough transform based on both cell counts and line-length ratios.;For high-level operations, a spatial-vector structure for efficient spatial search in line linking was developed. A best-neighbor finding function and four line-linking functions, each fully supported by the spatial-vector structure, were developed.;Vectorized road networks were overlaid on those from the TIGER/Line files on a GIS for evaluation. The results suggest that SPOT panchromatic images can be used to prepare and update road networks on rural and, to a certain extent, suburban areas where land cover is broad and new roads are likely to be added.