Detecting And Identifying Road Junctions From High Resolution Remote Sensing Images Based On DPM

With the rapid development of technology of satellites, spatial resolution of remote sensing images is constantly increasing, People obtained the remote sensing images from low-resolution to sub-meter level resolution, how to use these high-resolution remote sensing images, and extracted road junctions from them are becoming a new hot issue that domestic and foreign experts are studying actively.According to the characteristics of road network of high resolution remote sensing images, we defined it as a network system which the roads are interwoven each other. Road junctions which are formed by the influx of roads are the most important nodes of road network. If we can not acquire the position of road junctions, it will be difficult to construct the road network. Therefore, to identify the position of the road junctions is significant for improving precision of constructing road network, The traditional methods of extracting the road junctions is that we extract the roads firstly, and then the road junctions, which is complicated and the result of extraction is directly dependent on the accuracy and completeness of road extraction.The proposed method is based on deformable parts model. This is a method that we could extract road junctions directly. According to spectral, geometric features of road junction. It can be regarded as models with a certain deformation, whose parameters can be calculated by training samples and then the positions of intersections are searched by matching various models with images. The experiments indicate that the proposed method is able to achieve comparatively good performance in various road scenes, and that it can overcome the effects of interfering factors effectively, having better integrity and reliability.