Study On Urban Road Extraction And Surface Reconstruction By Combining Airborne Lidar Point Clouds And High Resolution Aerial Images

In recent years, while data acquisition competence in remote sensing earth observing was greatly improved, the corresponding data and information processing competence remains to be improved because it can not satisfy the increasing need for spacial information in all areas of national economy. A key scientific problem to be solved is how to extract target information of ground objects with high accuracy in the mass remote sensing data. As one of the main terrain factors of cities, the road is the main object studied in photography measurement and ground object information extraction. In the past40years, a large number of scholars have made semi-automatic or automatic researches on the road feature in terms of remote sensing image data. Although great development and renovation has been made both theoretically and technically, no practical approach for extracting urban road network has been adopted, and only some semi-automatic road extracting systems were developed. With the rapid development of3D digital city, people are having a higher and higher requirement for the resolution, accuracy and instantaneity of road information. Road extracting and reconstructing based on high resolution remote sensing date is the main trend of future development. Improvement in spacial resolution will greatly increase the number of non-road objects in images, bringing out greater challenges and difficulties to road extracting. As is shown by the existing researches and practice, this road extracting tact based on single data source cannot provide effective data, so it is sensible to combine more data sources.High resolution remote sensing image and airborne LiDAR point cloud data are complementary to each other in describing ground objects. High resolution remote sensing images are clear in edge but are often covered by shadow and trees. In contrast, the edge of airborne LiDAR point clouds is blurry of detail but the clouds are not affected by shadow. What’s more, its penetrability through vegetation and the3D information facilitate road modeling. The combination of these two types of data can satisfy the demand for extracting and reconstructing more accurate road features.The thesis researched on the key technologies in extracting urban roads and making3D surface reconstruction by combing high resolution aerial images and airborne LiDAR point cloud data. The main content and innovation points are as follows.1. The thesis looked into the whole technology roadmap for extracting and constructing the3D surface of urban roads with both airborne LiDAR point cloud data and high resolution aerial images. There are mainly three processes. Firstly, the author extracted the initial road central line network on the basis of airborne LiDAR point cloud. Secondly, he made automatic extraction of accurate2D road based on combined spoke wheel algorithm. Finally, he reconstructed the road surface with the improved self-adaptive butterfly subdivision surface algorithm. The research results showed the feasibility of this technology.2. The author researched on the approaches for extracting road centre lines based on airborne LiDAR and proposed the coarse-to-fine point cloud extracting strategy that combines the elevation, intensity of point clouds and geometrical features of roads. Based on this, he extracted road center lines with morphology and conceptual grouping. Firstly, he came up with a self-adaptive progressive encrypted triangle network filtering algorithm, which uses mobile curve surfaces, to gain the ground points. Then, he extracted the initial road point clouds out of ground point clouds based on the feature of ground point clouds in strength and came up with edge-length and area-threshold-value-based restricting Delaunay TIN approach, which takes the geometrical features of road into consideration, to make the initial road point clouds more accurate. Finally, he worked out the distance images based on refined road point clouds and extracted the complete road center line information with mathematical morphology and conceptual grouping.3. The author made key research on some key issues for extracting accurate road contour out of high resolution images with combined spoke wheel algorithm. The issues are as follows.(1) Automatic extraction of initial seed points in spoke wheel algorithm:With the initial road central line information extracted out of airborne LiDAR raod point clouds, the author made automatic initialization of the speed points for spoke wheel algorithm, thus improving the automaticity and accuracy of road extracting.(2) Eliminating the effect of shadowy road images with combined LiDAR featured spoke wheel:To solve the problem that the traditional spoke wheel algorithm based on the grayness of images is easily affected by shadow and noise and cannot come up with the road contour of the shadow region, the author made full use of the fact that airborne LiDAR point clouds are not affected by shadow and proposed to recognize the shadow region with the combined Fulan algorithm, which utilized the elevation features of airborne LiDAR road point clouds and the grayness of images. This algorithm repaired the lost information in the shadow region and then extracted the accurate road contour with the image grayness based spoke wheel algorithm. (3) Extracting double wheel road contour and eliminating the effects of double wheel road with combined LiDAR featured spoke wheel algorithm:Affected by the big difference in grayness, the traditional broke wheel algorithm cannot fully extract the contour of both parts of roads. In this thesis, the author extracted the road contour by automatically searching for the seed point of the other half of the road in the restricted area with the combined broke wheel algorithm. He came up with the integrated road contour and in this way, the wrong initial seed point position can be automatically adjusted.4. The author discussed the approaches to reconstruct road surface model by combining road contour with airborne LiDAR road cloud. Road surface model reconstruction is actually the process for expressing curve surface. The frequently used gridding and TIN surface modeling approach cannot effectively express the smoothness of the curve surface of the road, while subdivision surface algorithm can precisely describe it. Therefore, the author reconstructed the road surface with the self-adaptive improved butterfly subdivision algorithm and in the reconstructing process, he tried his best to describe the smoothness of curve surface of road while limiting the increase of data amount by calculating the triangles used in the self-adaptive subdivision according to the flatness of adjacent triangles.