Research On Key Technology Of Data Conflation Of Volunteered Street View Data And OpenStreetMap Data

In order to deal with the global climate change,countries around the world have put forward their own green development goals to reduce greenhouse gas emissions.In urban area,green commuter is one of the important means to achieve energy saving and emission reduction.Sustainable low-carbon transportation,mainly by walking and cycling,has gradually become an important mode of urban transportation in short and medium distance.There is a growing demand for digital map of non-motroized roads.However,the traditional road map only focuses on the expression of the geometric,topology and semantic information of motorized roads,but lacks representation of non-motorized roads.Therefore,OSM road data,which contains rich non-motorized roads information,was used as map data source by non-motorized road users.However,due to the fact that OSM data is generated by a large number of untrained volunteers,the data production process lacks strict quality control means,resulting in data incompleteness and other quality issues.As a novel Street View data,Volunteered Street View data can cover non-motorized road scenes such as parks,mountains and bikeways that are inaccessible to commercial Street View data-collection vehicles.The GPS information of Volunteered Street View data can reflects the morphological structure of the road network in the real world.At the same time,the captured images contain a lot of semantic information related to road networks.In this paper,Volunteered Street View data is used to extract bikeway road networks and its corresponding attribute information.The data conflation framework of extracted road networks and OSM road networks is analyzed,in order to provide a more complete data source for bikeway road networks,theoretical and technical supports for data mining and data fusion methods of heterogeneous geographic crowdsourcing data.The research content of this paper mainly includes the following aspects:1.The basical cocepts of Volunteered Street View data are introduced.After the introduction of the generation history of Volunteered Street View data,the differences between the Volunteered Street View data and commercial Street View data are compared,and the advantages and disadvantages of the two types of data are analyzed.Then,the current state of Volunteered Street View data is introduced in detail.Specifically the data growth trends,data coverage and user community activities of Mapillary are analized.2.The framework of extracting bikeway road networks from Volunteered Street View data is proposed.Firstly,the classification method of Volunteered Street View sequences based on transportation mode is proposed.Movement features and geometric features are extracted from GPS trajectories,and tree ensemble methods are used as classifers to indentifiy transportation mode of trajectories based on extracted features.Then,a raster-based road network extraction method is proposed to extract bikeway road networks from the classified sequences.3.To solve the issue of attribute incompleteness of OSM road data,this paper proposes to extract traffic sign and road marking from Volunteered Street View images as road attribute information.Firstly,the basic theoretical knowledge of deep learning is introduced,and the advantages of deep learning method compared with traditional methods are expounded.Then,the current mainstream object detection models based on deep learning are introduced.Based on this,the improved YOLOv3 model and the shallow classification model for traffic sign detection and recognition are proposed.Finally,the extracted attribute information is assigned to the generated road network through map matching method.4.In view of the difference in the level of detail between OSM road network and generated road networks,which leads to the difficulty of road network matching,corresponding solutions are put forward.Firstly,a polygon-based multi-lane road detection and simplification method is proposed.Based on the extracted shape and topological features of the multi-lane road polygon,the tree based ensemble method is used to identify multi-lane road recognition in OSM road data.Then,the detected multi-lane road polygons are simplified to the centerline of roads through the raster-based method.After pre-processing,the identification of corresponding road objectes between the generated road network and the simplified OSM road network is conducted based on map matching methods.