Research On Road Network Rapid Extraction Method Based On Vehicle Trajectory Data

Urban traffic construction is an important part of urban development.It is closely related to people's daily life.With the rapid development of electronic communication technology,Internet technology and Internet of things technology,urban road network data have gradually become more and more important in urban planning,road supervision and map value-added services.No matter the expansion of the city and the optimization of its internal structure,it may lead to the change of the urban road network.The traditional urban road network data are mainly obtained by two ways,professional personnel mapping and remote sensing image digitization.However,the two methods have high investment cost,high professional technical requirements and long production cycle,so it is difficult to meet the requirements of the rapid development of the city for the network data.With the development of navigation and positioning technology(outdoor BEIDOU,GPS;indoor Bluetooth,WLAN and inertial positioning,etc.),wireless transmission technology and electronic communication technology,a large amounts of trajectory data are produced,such as cell phone signaling data,indoor location data and floating car trajectory data.Among them,the floating car trajectory data is generated when the vehicle is running on the road,so it implies rich road information.With the gradual opening of floating car trajectory data,it is possible to extract road network information from it.Many scholars and experts have done a lot of research on the extraction of road network information from floating car trajectory data and have made some valuable research results.But most algorithms extract road network,using a unified threshold to ignore the density difference of the trajectory data,and only consider the shape of the trajectory without considering the direction of the trajectory,which seriously affects the geometric accuracy and topological correctness of the extraction results.For this purpose:(1)two typical algorithms(raster-based road network extraction method and road network extraction method based on irregular constrained triangulation network)are reproduced and improved.Among them,a new raster method is proposed for grid-based road network extraction,which enhances the connectivity of the grid image and overcomes the difficulty of extracting the sparse section of the trajectory point because of the large difference of the trajectory point density.At the same time,the NL-means algorithm is introduced to optimize the grid image denoising,effectively overcome the effect of noise on the extraction of road network,and a new road polygon extraction method for road network extraction method based on constrained triangulation network is proposed to overcome the problem that the extraction threshold is difficult to set.Road centerline extraction,using Fermat point connection method design more reasonable intersection.In addition,a new optimization method is designed for intersection merging and fine edge elimination,which improves the reality of road network extraction.(2)an adaptive radius centroid drift clustering method is proposed,which can automatically adjust the clustering parameters according to the trajectory density and road width and realize the road topology connection by using the path direction.First,the road network skeleton points are calculated by the adaptive radius centroid drift clustering method,and the wavelet clustering algorithm is used to obtain the direction set of the skeleton points of the road network.Then,the skeleton points are recursively connected according to the radius and direction of the clustering,and the road network data are generated.The experimental results of the floating car trajectory data of one day in Futian District,Shenzhen,are tested and verified.The experimental results are qualitatively and quantitatively evaluated with the grid method and the constrained triangulation method.The experimental results show that the extracted road network data in this algorithm is significantly improved in geometric accuracy and topology accuracy,and is suitable for large data processing.