| The digital maps provide road information for Unmanned Ground Vehicles(UGV)positioning,layout estimation and holistic scene understanding.It is an important part of UGV’s self-decision and global path planning.Most current digital maps are still inadequate for advanced applications in unstructured environments.Lack of up-to-date information and detailed road networks is among the most important shortcomings.This cannot meet the need of UGV’s autonomous navigation.Therefore,we need to design an efficient map-building method for UGV.Most UGVs rely on road networks of the detailed maps of the environment to facilitate navigation and perception.However,the roads in the map can not be applied to UGV’s navigation directly,while the method of manual intervention limits the applicability of current approaches.It is necessary to extract road topological networks from the built map to improve the capabilities of UGV’s autonomous navigation and planning.In order to solve the problem,this paper aims to generate a two-dimensional or threedimensional map from a serial of images and to extract the road area,so that UGV can accomplish the task of planning based on prior information.In this paper,the construction of the 2D and 3D map and the extraction of the roads from a 2D image are described.1.The paper presents a method of building a 2D map using panoramic images.The panoramic camera and the high-precision positioning system are mounted on the unmanned vehicle,which can acquire the panoramic images and the corresponding vehicle positioning information.The panoramic image is converted from the spherical coordinate system to the vehicle-based coordinate system to obtain an overhead view centered on the unmanned vehicle,which is called inverse perspective mapping(IPM).And then a series of IPM images are spliced into a large-scale 2D map according to the positions from high-precision positioning system.Based on the map,topology network of 2D map is constructed after adding a variety of attributes of roads in the GIS database.The map is the basis for unmanned navigation and decisions,offering the road and other prior information.2.A method is presented to reconstruct the 3D structure of geospatial space from aerial images using multi-view geometric.It includes feature extraction from the aerial image datasets,and feature matching of different images according to the feature points.Epipolar constraint is adopted to remove false matches so that the unordered datasets become ”orderly”.Afterwards,the two images with the largest number of matching feature points are reconstructed to obtain the initial point clouds with the multi-view geometry algorithms.And the rest of the images are incrementally added to obtain the camera poses of all the images and the sparse point clouds of the entire geographic space.In the process of optimization for 3D points and camera poses,the paper not only minimizes the re-projection error of 3D points,but also minimizes the distances between the camera centers and the GPS positions.The optimization operations refine the camera poses and sparse point clouds composed of 3D points to decrease the error of models.The dense point clouds are obtained by calculating the depth-maps and merging them,which can be applied to UGV’s navigation and planning.3.Based on the above work,a method of extracting roads from 2D images is proposed.Firstly,a serial of road areas in the image are chosen as the training sample which the least square method is applied to.And then spectral features of the training sample are fitted and applied to the other sample to obtain the road binary image.Secondly,open and close algorithms are adopted to smooth the binary image.And the method of polygon detection is designed to remove the interference of non-road areas,such as house edges.Then the road binary image is converted into a road network composed of single pixels through topological thinning.Finally,the final topology of the road network is gained by removing the short branches in the road network.4.In order to validate the effectiveness of the proposed algorithm,the experimental platforms for map construction and road extraction are set up.The efficiency and accuracy of the algorithm are verified by using multiple datasets.Firstly,the effect of the 2D constructed map on the UGV’s global path planning is demonstrated through experiments.By using two image datasets,we can reconstruct them to 3D geospatial maps and analyze the results of experiments to get the accuracy of the map-building method.Finally,the precision of the road topology is calculated.The validity of the proposed method in different scenarios is verified by experiments. |
