Research On Time-sensitive Trajectory Repair

In recent years,trajectory data mining has become a hot topic.A large number of people have participated in this research since Zheng Yu et al.systematically proposed the concept of Urban Computing in 2014.Nowadays,with the development of intelligent transportation,traffic information acquisition sensors are spread all over the country.It is more and more convenient to acquire spatio-temporal data of geographical markers through sensors,so the volume of the traffic data grows rapidly.Those data make it possible to do researches of traffic trajectories mining.The bayonet data is an important type of traffic data.The data are collected by high-resolution cameras distributed on the road.Whenever a vehicle passes by a camera,the camera records the traffic of the vehicle.However,due to equipment sensitivity,inaccurate recognition,and equipment damage,problems such as geographical location jumps and data loss often occur.Such data not only affects the real-time control of traffic,but also reduces the performance of further data analysis.To solve this problem,this paper proposes a Bi-directional Time-sensitive Markov Model(TBiMM)for predicting missing points of trajectory data.Markov model,a statistical model,is often used for the prediction of spatio-temporal data.The traditional Markov model only considers unidirectional information,that is,the preceding data or following data relative to the point to be predicted.The data used in this paper are the whole trajectory data.The missing points are in the middle,so both the preceding data and the following data are known,and the information that can be used is more comprehensive.Therefore,we use the bi-directional Markov model,we consider four Markov models with different combinations of directions,which are RR,RL,LL and LR,to comprehensively consider the bi-directional information of the missing points.Time is an important factor in the trajectory data.In order to include time information,we combine the position information of a record with its corresponding time period as a state of the Markov model.To quickly retrieve states,a state indexing structure is proposed in this paper.The binary form of a location ID is spliced together with the binary code of its corresponding time bin.The decimal number of the spliced binary code is the state ID.In this representation,we can also index location ID and time bin number by state ID backward.By using this state indexing structure,the TBiMM is finally proposed,the model can not only predict the location of missing points,but also predict the time period during which missing points occur.This paper compares TBiMM with other baseline models on real data sets.The experimental results show that the model proposed in this paper effectively improves the accuracy of prediction and can effectively predict the time period when missing points occur.The main contributions of this paper are as follows.(1)A state indexing schema is proposed,which combines time and location information together.We can retrieve the state from time and location and retrieve the time and location from the state reverse.With the support of this structure,the model in this paper can not only predict the location of missing points,but also obtain the time period for generating missing points.(2)Utilizes four different Markov models to consider two-way information of missing points synthetically and the votes mechanism is used to combine the decision results of the four models.(3)The trajectory data sets with different missing rates are simulated on real data sets,and we compare the performance of each model on these data sets.Experimental results show that the proposed model TBiMM is superior to other baseline models.