Key Techniques Of Trajectory Data Analysis And Processing

In recent years,people's demand for transportation and travel has become stronger and stronger,intelligent transportation research has become a hot topic.Efficient intelligent transportation can bring convenience for people and improve driving safety and transportation efficiency.However,with the development of economy and population,it also brings us some problems and challenges: 1.There are inherent errors and random errors between the geographic locations of the trajectories collected by using the Global Position System(GPS)and the actual geographical locations,furthermore,with the appearance of more and more low-sampling-rate trajectory data,the difficulties of restoring the real trajectories are further increased.2.The urban congestion has become a serious problem with the appearances of more and more urban cars,how to utilize existing vehicles to improve potential capacity for transporting people is of great significance to alleviating urban congestion,reducing passenger travel costs and increasing driver income.3.People have become increasingly concerned about the accurate travel time estimation as the pace of people's life continues to accelerate,how to deal with massive,complex and multi-class data that affects travel time,how to provide accurate time estimation method for people so that travellers can plan routes in advance,save travel costs and reduce the travel blindness.The main research contributions of this paper are as follows:1.Propose a map matching method based on Hidden Markov Model by using multi-factors.The method comprehensively considers the distance between the sampling points and the candidate points,the angle between the sampling segments and the candidate segments,the vehicle speed and the road speed limit,and the road network topology factors,and proposes a convex function that uniformly describes the transition probability,and use the fast Viterbi algorithm to select the optimal candidate path.The experimental results show that this method can effectively reduce the errors between the sampling positions and the real positions,and lay a foundation for other trajectory research.2.Propose an effective dynamic multi-passengers ride-sharing model.A grid-based road network partition method is introduced to reduce the time for calculating the shortest path and the shortest distance.In order to select the optimal passenger,a path similarity approach which takes in both distance and time into consideration is designed.A multi-branch tree is introduced to combine the pick-up and delivery sequence of drivers and passengers.In order to improve operational efficiency,a two-step tree pruning method is presented to reduce the possible solution searching scope.The experimental results show that the proposed model can effectively solve the dynamic multi-passengers ride-sharing problem,improve the driver sharing path ratio,reduce the passenger waiting time,and raise the overall utilization rate of vehicle resource in the city.3.Propose an end-to-end travel time estimation method based on deep learning.The method jointly employs CNN and LSTM to extract spatial and temporal features of trajectories respectively and introduces an attention mechanism for judging difference of the road segments weight influences on travel time.In addition,external factors such as weather,time and types of roads are considered comprehensively and they are integrated into the training model.Experimental results show that the proposed method is significantly better than some current advanced benchmark methods and is more robust.