With the rapid development of high-resolution remote sensors, how to rapidly, automatically and accurately extract road intersections using high-resolution remote sensing images has become a focus which draws attention in the academic field and several management departments. The current methods of road networks extraction can be divided into two types. The first kind method is to construct road networks according as road intersections computation on the basis of road extraction, but it exists the defect and deficiency of full automatic and suitability. The second kind method is to construct road networks according as road extraction on the basis of road intersections extraction, but it is almost appropriated for low-resolution remote sensing image. Therefore, in order to solve the problem of geometric position detection and geometric shape type identification of road intersections in high-resolution remote sensing images, this paper proposes a road intersections detecting method based on the deformable part models:Firstly, the image features of road intersections are studied in high-resolution remote sensing image; Secondly, road intersection is regarded as a object which is made up of root and many parts, meanwhile, the model parameters of road intersections are obtained through feature extraction and model training using the deformable part models; Finally, the geometric position and geometric shape type of road intersections are obtained using sliding windows matching method. The result of simulation and experiment show that the new method can not only automatically and accurately detect the location of road intersections, but also achieve the geometric shape type of road intersections. Research in this method will effectively improve work efficiency of road network topology construction.