Vehicle Trajectory Path Restoration Method Based On Sparse Taxi GPS Data

For the fine-granularity vehicle trajectory data,we can realize the traveler path judgment by matching the traveler position with the road network,and further get the characteristics of the traveler’s path selection behavior.However,for the data with high missing rate of trajectory data,the density of data points mapped to the road network decreases,if the distance between adjacent GPS points is far,or even there are multiple intersections,then the selected path will have many possibilities,it is more difficult to match the road network,and it is more difficult to restore the real route of the vehicle.In view of this situation,the spatio-temporal trajectory of the sparse taxi GPS data travelling in every street of the Ningbo city is taken as the research object.We propose a sparse GPS trajectory path restoration method based on Bayesian network.Different from the traditional research based only on space-time variables,this method also considers the characteristic factors such as operation,environment,driver,traffic and so on.By incorporating multiple correlation factors and discretizing the data,we established a Bayesian network model of vehicle path restoration with a view to improving the prediction accuracy.Taking the GPS trajectory data of Ningbo taxi service information management platform as the test object to verify the applicability of the method.The case results show that the Bayesian network method based on multiple factors is superior to the Logit selection model that the prediction accuracy of path choice reaches 91.2%.In addition,the proposed method is especially suitable for scenarios with a high missing rate of track data,such as a scene with missing track point span of about 5 min.In order to study the effect of the missing rate of the track data on the prediction model established in this paper,the research scope is expanded,the modeling and the calculation precision are carried out again,the accuracy of the model is predicted by the ROC curve,The travel times between origin and destination are around 8min,12 min,15 min respectively.After choosing these data missing rates,the AUC value is 0.825,0.787 and 0.56 separately.It confirmed in the case of the research of this paper this method is suitable for the data loss rate about 5 min.However,when the data loss rate reaches 15 min,this method does not have obvious advantage.