Research On Trajectory Recovery Based On Deep Learning

With the extensive use of satellite navigation and location-enabled devices,the trajectory of a moving object can be captured by multiple information systems and stored in multiple databases.Trajectory recovery is to discriminate trajectory information belonging to the same moving object in multiple databases,in order to recover and reconstruct a relatively complete trajectory of the moving object.For the problem of trajectory recovery,a method based on RNN ensemble learning is proposed,which aims at the different sparse degree of the sampling points.Firstly,the formal model of trajectory recovery is defined,and then each train set is divided into multiple train subsets by using sampling point time sparsity evaluation function.Next,RNN network model is used to describe the degree of splicing of trajectories from different databases.Finally,an ensemble learning strategy is used to integrate multiple RNN networks to achieve the goal of recovering trajectories of moving objects from multiple databases.Another trajectory recovery method is based on RNN-SVM,considering the connection features between the trajectories contained in the connection pairs.The method is modeled according to the trajectory continuity,density and time-series distribution features of connection pairs.This model automatically acquires the time-series distribution features of the connection pairs through the RNN,then uses this feature,continuity and density for SVM classification to obtain trajectories related to moving object,in order to achieve the goal of trajectory recovery.Experiments show that the trajectory recovery ensemble learning method and RNN-SVM trajectory recovery method can capture the spatio-temporal continuity features and the trajectory connection features respectively,and both can search related trajectories from different databases with higher accuracy.They achieve the goal of recovering the trajectory of the moving object in different databases.