Road Extraction And Vehicle Detection Of High Resolution Remote Sensing Images

With the development of high resolution remote sensing technologies, more and more texture details appear in the images, bringing much opportunities as well as challenges for the information extraction. As an important surface feature, the extraction of road information has broad application prospect and value of research. However, the road information displays differently between the high resolution images and the low resolution images, resulting in that some old methods are not applicable. This article deals with the road extraction method based on segmentation using three data sources(GF-1, GF-2, and BJ-2). In addition, the high resolution remote sensing images reveal more details of the roads, making vehicle detection possible. This article makes some improvement based on the double-threshold vehicle detection methods. The main research includes:First of all, this article focuses on the road extraction method based on the clustering segmentation. The fuzzy-c-means segmentation and the K-means segmentation are chosen to deal with the images. After clustering, some other objects which have similar spectral features with roads such as architectures and so on are extracted. As a result, post-processing is made using mathematical morphology and shape filter.The roads in mountain area are so long, thin and winding that some breaks always appear with the shadows of tree. In addition, the radio of break distance and width of the mountain road is relatively larger, which makes the traditional methods out of use. This article raises a new method to connect the broken road segments based on the distance threshold of convex hull and brings a good result.Following that, this article focuses on the road extraction using the improved watershed algorithm. Foreground makers based on the adaptive thresholds and backgrounds markers based on the morphology reconstruction are used to modify the gradient image, reducing the over-segmentations. After getting the segmentation result, shape and spectral information is used to extract the roads.Finally, the article detects the bright and dark cars on the road using the improved double-threshold methods and removes disturbing objects as well as shadows. To solve the problem of false detection caused by the dark windows on bright cars of GF-2 images, a new method which is proved to be useful is raised to detect the integrity of cars.