On Navigation Geographic Information Extraction Method Based On Massive Vehicle Trajectory Data

In the era of big data and artificial intelligence,people have an increasingly strong demand for comprehensive,accurate and real-time navigation geographic information.The traditional navigation geographic information acquisition methods are mainly based on field measurement,traverse measurement system "sweeping the street",aerial survey and remote sensing.The first two acquisition methods have a long data acquisition period and limited coverage,and the subsequent processing work is large and costly.Although the latter two acquisition methods have a wide coverage and short data acquisition period,they have technical difficulties and limited information extraction,which seriously restricts the timely collection and update of navigation geography information.At present,there is still a lack of simple,economical and practical new technologies at home and abroad that can timely acquire and process such information.Therefore,how to collect or automatically extract navigational geographic information with high accuracy and strong current situation has become an important problem to be solved.In recent years,the advancement of positioning technology and the price of positioning equipment have dropped drastically.Many mobile devices such as automobiles,mobile phones,and PADs have installed GNSS receivers,and almost continuously generate massive data(called trajectory data)with time and position coordinates.Through these trajectory data,it is possible to reconstruct the moving trajectory of the object,and extract the hidden road geometric position,traffic rules,road network POI and real-time road condition information,which is of great value.These trajectory data sources have high coverage,stable source,large scale,low cost and newly updated,which provides an unprecedented opportunity for large-scale automatic extraction of navigation geography information.Based on large-scale taxi trajectory data,this paper deeply studies the key technologies of navigation geographic information extraction for navigation electronic map production,realizes geometric information extraction for conventional roads and complex intersections,typical road network POI identification and travel time prediction of the whole road network.The results of the thesis can provide a feasible technical reference for efficient,cheap,accurate and real-time navigation map production and update.The main research contents and conclusions of this paper are as follows:(1)The current status of trajectory-based navigation geographic information extraction is reviewed.It mainly includes the research of geometric road network information extraction method,road network POI information identification research and road network travel time estimation and prediction method,and summarizes road network geometry.The main problems to be solved in the four aspects of skeleton construction,geometric intersection information extraction,road network POI identification and travel time prediction are summarized.(2)In view of the problem that the road network under the current large-scale rough positioning trajectory data is automatically constructed without considering the road complexity difference,a road network incremental construction algorithm considering the road network complexity is proposed.Based on the road driving rules,the regular sub-tracks are selected from the rough trajectory data,and the complexity of the road network around the point to be processed is calculated based on the information entropy,and the merge parameters are automatically adjusted accordingly,and the road network center line is incrementally constructed.The experimental results show that the algorithm can adapt to the complex environment of the road and effectively extract the traffic rules information of the road network skeleton.Compared with the typical algorithm of the same kind,the algorithm constructs the road network to reduce the total number of road network nodes and improve the fineness of the road network expression.(3)Aiming at the problem that the current geometric information extraction of complex intersections is inaccurate and difficult,a multi-level complex intersection recognition method based on mathematical morphology is proposed.The massive trajectory data in the intersection area is transformed into the density bitmap of the road coverage,and the noise is removed by the filtering operation.By setting a plurality of gray thresholds,the multi-level road center lines are extracted by the morphological method,then geometric refinement and the topological information extraction such as the road direction were performed.The experimental results show that the proposed method can improve the accuracy of complex intersection geometry and steering rule information extraction,and the algorithm has improved extraction accuracy compared with similar algorithms.(4)Aiming at the problem that the recognition rate of road network POI based on feature engineering method relies heavily on the selection of artificial features and the recognition rate is not high,a typical road network POI recognition method based on convolutional neural network is proposed.By analyzing the spatiotemporal pattern of trajectory in the typical road network POI area,the trajectory expression model suitable for convolutional neural network input is proposed,and different network structures and hyper-parameters are explored to realize the optimal road network POI recognition convolutional neural network model.The experimental results show that the method can effectively identify the typical road network interest points such as traffic lights,gas stations,and compared with similar algorithms,the algorithm improves the typical road network POI recognition rate.(5)Aiming at the problem that the empirical rule method and statistical method are not suitable for large-scale complex road network travel time prediction,a travel time estimation method based on graph convolution LSTM is proposed.The road network is formally expressed,and the road network travel time convolution calculation is defined,then the travel time prediction model combining Long Short-Term Memory is completed.The experimental results show that the method can accurately predict the travel time of the whole network in different time periods,and compare with similar algorithms,the prediction accuracy of the algorithm is greatly improved.