Automatic Detection of Changes in Road Databases Using Satellite Imagery

An automatic procedure is developed for detection of changes between an existing road database and a newly registered satellite image, rectified to an orthophoto. The approach uses the existing database to force the detection and delineation of the corresponding road network in the image. Each road segment is handled separately during the comparison between the image and the database as the position errors are considered to be randomly distributed along the road network. ”Ziplock snakes” are used to optimise the delineation process. The algorithm minimises, gradually, the internal and external energy sum represented by road bending respectively an eigenvalue of each image pixel calculated from the Hessian matrix formed by the partial derivatives of a Gaussian filter. Logically new roads are connected to the existing network. Statistics from the image roads in combination with line linkage procedures form the method that generates the new roads supplementing the detected image roads. Finally the comparison between the old road database and the roads detected in the image is performed. Changes are signalled if exceeding limits calculated from the variances of geometrical accuracy of the roads, internal image accuracy and the mathematical model of the orthophoto production.