In this paper a semiautomatic method is proposed for road extraction by combining a stereoscopic pair of low-resolution aerial images and a polyhedron generated by using a digital terrain model (DTM). The problem is formulated in the object space by means of an objective function which models the object 'road' as a smooth curve that belongs to a polyhedral surface. The proposed objective function also depends on the radiometric information, which is accessed in the image space via collinearity relation between road points, in the object space and corresponding points in the image spaces of stereoscopic images. The polyline providing the best representation for a selected road is obtained in the object space via optimization of the objective function by using the dynamic programming algorithm. The optimization process is iterative and an operator needs to supply an initial polyline approximating the selected road. The obtained results showed that the method is robust against anomalies existing along roads, such as obstruction caused by shadows and trees.
Dynamic Programming; Road Model; Feature Extraction; Road Extraction; Aerial Images
