Automatic Road Extraction In Urban Areas Based On LiDAR Data And Remote Sensing Image

The automatic extraction of roads in complexs urban areas is urgent needs of society and industries. There are many strategies and methods have been proposed for road extraction from different sensor datas. However, roads extraction is a very difficult task due to the occlusions and shadows of the contextual objects and the complex pattern of the roads.Road extraction only from high resolution images in urban built-up areas is difficult due to the complexity of the scene, such as rich details of road in radiometry and contextual objects, like moving vehicles, and lane markers etc. Buildings can occlude roads and cast shadows on the road surface. The occlusion and shadows make road detection, tracking and forming of the road network more difficult in urban areas than that in rural areas. Using more information for the road extraction in the urban environment is a key to reach the goal of high correctness and completeness. LiDAR delivers accurately georeferenced set of dense point clouds with the intensity of the returned signals. Compared to aerial imagery, LiDAR has both advantages and disadvantages with respect to automatic object extraction. LiDAR directly provides3D points with less occlusions and smaller shadows. Surface roughness can easily be obtained from LiDAR, and is a distinguishing feature of man-made objects. However, For LiDAR data, there is no additional scene information directly available from a single point. The spatial resolution of LiDAR point cloud can not match the resolution of aerial imagery yet. In contrast to laser points, aerial images contain massive scene information and can reach high spatial resolution. Therefore, properly combining LiDAR data and imagery can improve the performance of automatic road extraction.In this paper, the extraction of road center lines based on LiDAR data are researched and explored, as well as the optimization of road network by the multi-information fusion from LiDAR data and imagery. The main contents are as follows:1. The domestic and overseas researches status from three perspectives are summarized in this paper. They are road extraction based on imagery, LiDAR data, and the two data sets, respectively. Main methods of road extraction are also summarized.2. The extraction strategies of road center lines from imagery are researched, which inlude the segmentaion and classification of road regions by the spectral and shape features, the detection technology of road centerlines from ribbon road areas, the primitive grouping method.3. A new method of automatic road extracion from LiDAR data are proposed. This method is to effectively to separate connected non road points from ground points, and extract road center lines without iamge interpolation and prior model. The non road points connected with roads contain vegetation, parking lots and bare ground. The method does not only eliminating the need for image smoothing, morphological method, tracing, thinning, and other operations, but also reduce large number of parameters setting.There are three steps of the method:road center lines detecting using adaptive mean shift clustering, the salient linear features enhancing bu stick tensor voting, road primitives extracion by weighted Hough transform.4. An construction and optimization method of road network from imagery and LiDAR data are researched. An approach of multi-information fusion from the LiDAR points and imagery is used to improve the correctness of the road center lines and topological completeness. Intersection with connected road direction are designed and extraced for road network construction. A analysis method of intersection contour is proposed to verify the information of intersection. The particle swarm optimization and dynamic programming are used to optimize the road network, which is considered as a multi-path decision optimization problem. The cost function of road models is built by multi-information.5. Several data sets are used to validate the road extraction method from LiDAR data proposed in this paper. The proposed method which are compared with the three other methods, produces significantly better correctness and quality. The method of road network optimization also improves the correctness of extraction results.