Integration method of profile matching and template matching for road extraction from high resolution remotely sensed imagery

In this paper, a novel tracker for semi-automatic extraction of ribbon road centerlines from high resolution remotely sensed imagery is proposed. Actually, our approach is an integration of least squares profile matching and least squares rectangular template matching. After initialization, a road template model is built which is composed of two parts: a profile perpendicular to the road axis, and some rectangular templates of strips of road marks or strips of vegetation parallel to road moving direction. In tracking process, least squares matching is employed to search road centerline points, and parabola is deployed to model the road trajectory to predict the position of subsequent road points and to guide the tracking go through bad road conditions. Extensive experiments demonstrate that our proposed algorithm can fast and reliably trace roads with road marks or strips of vegetation despite of appearance of much occlusion from trees, building or vehicles.