Research On Road Extraction From Remote Sensing Images Based On Level Set Method

The road character is very important in the basic geographic information. Updating the road network in time is meaningful to traffic management, urban planning, automatic vehicle navigation, and emergency services. With the development of RS technology, the quality of high-resolution satellite images improve, access cycles shorten. It provides the probability of extracting the road from the remote sensing (RS) images with the computer technology. How to extract road information from RS images draw increasing attention.Level set method is a class of curve evolution methods based on geometric active contour model without parameters. It is capable of modeling to arbitrarily complex shape, and handling the topological changes like a split, merger. This makes the level set method highly valuable in the engineering application.The works in this paper are focus on the application of level set methods in road extraction from the RS images. The main contributions and innovations are as follows:(1) An improved level set segmentation model without re-initialization based on multi-color space is constructed. The traditional level set methods for image segmentation are computational expensive due to the re-initialization. The proposed approach introduces a distance regularizing term into the CV model to keep the level set approach on the signed distance function. Then, according to the characteristics of high-resolution RS images, the proposed approach makes full use of gray value and each channel of HSI model. The proposed method was tested in RS images. The impact of the weight ratio of gray value and the mixing value of the HSI channel in the improved model on the extraction results was discussed. Finally, the completeness and correctness was used to evaluate the presented approach. The results show that the improved method segments the road area completely and accurately.(2) An automatic road extraction method is proposed and a framework of dividing large images into sub-images is designed based on the road centerline relay guidance. For the location of the initial curve may result in failure, an adaptive dual-threshold segmentation combined with morphological operations is developed to obtain the initial curve position and extract road automatically. Because of the large size of RS images, a framework of dividing large images into sub-images is designed based on the road centerline relay guidance. That is utilizing the tilt angle of road centerline in current sub-image to get the location of next sub-image. Results show that the initial curves lie closely to the road contour, which make automatic road extraction based on the level set happen; the framework joining with the level set method eliminates the background interference obviously.(3)Software for extracting roads from high-resolution RS images is developed, and some basic functions such as an area filter, road centerline extraction, adaptive threshold algorithm, morphological expansion and corrosion, curve correction algorithm is packaged into a DLL dynamic link library. The software contains the image pre-processing functions and the road feature extraction functions. It divides the large RS image into sub-images, then extracts the roads from each sub-image with the level set method, finally puts them together. The software has passed the test of Zhejiang software testing center.